CN112085104B - Event feature extraction method and device, storage medium and electronic equipment - Google Patents
Event feature extraction method and device, storage medium and electronic equipment Download PDFInfo
- Publication number
- CN112085104B CN112085104B CN202010946188.4A CN202010946188A CN112085104B CN 112085104 B CN112085104 B CN 112085104B CN 202010946188 A CN202010946188 A CN 202010946188A CN 112085104 B CN112085104 B CN 112085104B
- Authority
- CN
- China
- Prior art keywords
- event
- network model
- internal
- neural network
- events
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 39
- 238000003062 neural network model Methods 0.000 claims abstract description 62
- 238000012545 processing Methods 0.000 claims abstract description 33
- 230000004927 fusion Effects 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 30
- 239000011159 matrix material Substances 0.000 claims description 38
- 238000012549 training Methods 0.000 claims description 37
- 238000010586 diagram Methods 0.000 claims description 32
- 230000006870 function Effects 0.000 claims description 24
- 230000002776 aggregation Effects 0.000 claims description 9
- 238000004220 aggregation Methods 0.000 claims description 9
- 230000009471 action Effects 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 10
- 238000005516 engineering process Methods 0.000 abstract description 7
- 238000004891 communication Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000007480 spreading Effects 0.000 description 2
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Machine Translation (AREA)
Abstract
The application provides an event feature extraction method, an event feature extraction device, a storage medium and electronic equipment. Firstly, taking an event text as input of a node neural network model, outputting corresponding internal features by the node neural network model, taking an event theme as input of an associated neural network model, and outputting corresponding external features by the associated neural network model according to a relational network; taking the internal features and the external features as inputs of a fusion network model, and outputting event features corresponding to the events by the fusion network model; the internal characteristics of the event are concerned, and the fact that massive text information and external characteristics among a large number of events can be derived in the process of generating and transmitting the event is considered as the network science and technology develop. And the external characteristics and the internal characteristics are fused, so that the obtained event characteristics are more perfect, and the subsequent data processing is more convenient.
Description
Technical Field
The present invention relates to the field of data processing, and in particular, to an event feature extraction method, an event feature extraction device, a storage medium, and an electronic device.
Background
With the development and application of the internet, big data is widely applied to various fields. Event analysis is one of the important branch trends. An event refers to a change in state that occurs at a particular point in time or time period, within a particular geographic area, and that consists of one or more actions that one or more characters are engaged in. The basis of event analysis is to extract event features.
In the existing event feature extraction method, only the internal features of the event, such as people, places, time and the like in the event, are often concerned. Neglecting that events derive massive text information during the generation and propagation process with the development of network science and technology, external features among a large number of events may be derived. This is also a technical problem to be solved by the person skilled in the art.
Disclosure of Invention
An object of the present application is to provide an event feature extraction method, an event feature extraction device, a storage medium, and an electronic device, so as to solve the above-mentioned problems.
In order to achieve the above purpose, the technical solution adopted in the embodiment of the present application is as follows:
in a first aspect, an embodiment of the present application provides an event feature extraction method, where the method includes:
taking an event text as input of a node neural network model, and outputting corresponding internal features by the node neural network model, wherein the internal features are feature representations generated by combining all internal nodes in an event, and the internal nodes are words in the event text;
taking an event theme as input of an associated neural network model, and outputting corresponding external features by the associated neural network model according to a relational network, wherein the event theme corresponds to the event text; the relation network comprises the event and other events which have direct or indirect relation with the event, and the external characteristics are characteristic representations of the association degree of the event and the other events;
and taking the internal features and the external features as inputs of a fusion network model, and outputting event features corresponding to the event by the fusion network model, wherein the event features are integral feature representations of the event.
In a second aspect, an embodiment of the present application provides an event feature extraction apparatus, including:
the processing unit is used for taking the event text as the input of a node neural network model, and outputting corresponding internal features by the node neural network model, wherein the internal features are feature representations generated by combining all internal nodes in the event, and the internal nodes are words in the event text; the method is also used for taking the event theme as input of an associated neural network model, and the associated neural network model outputs corresponding external characteristics according to a relational network, wherein the event theme corresponds to the event text; the relation network comprises the event and other events which have direct or indirect relation with the event, and the external characteristics are characteristic representations of the association degree of the event and the other events;
and the fusion unit is used for taking the internal features and the external features as the input of a fusion network model, and outputting event features corresponding to the events by the fusion network model, wherein the event features are integral feature representations of the events.
In a third aspect, embodiments of the present application provide a storage medium having stored thereon a computer program which, when executed by a processor, implements the method described above.
In a fourth aspect, an embodiment of the present application provides an electronic device, including: a processor and a memory for storing one or more programs; the above-described method is implemented when the one or more programs are executed by the processor.
Compared with the prior art, the event feature extraction method, the device, the storage medium and the electronic equipment provided by the embodiment of the application have the beneficial effects that: firstly, taking an event text as input of a node neural network model, outputting corresponding internal features by the node neural network model, taking an event theme as input of an associated neural network model, and outputting corresponding external features by the associated neural network model according to a relational network; taking the internal features and the external features as inputs of a fusion network model, and outputting event features corresponding to the events by the fusion network model; the internal characteristics of the event are concerned, and the fact that massive text information and external characteristics among a large number of events can be derived in the process of generating and transmitting the event is considered as the network science and technology develop. And the external characteristics and the internal characteristics are fused, so that the obtained event characteristics are more perfect, and the subsequent data processing is more convenient.
In order to make the above objects, features and advantages of the present application more comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered limiting in scope, and that other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic structural diagram of an electronic device according to an embodiment of the present application;
fig. 2 is a flow chart of an event feature extraction method according to an embodiment of the present application;
fig. 3 is a schematic diagram of sub-steps of S105 provided in an embodiment of the present application;
fig. 4 is a schematic diagram of the substeps of S106 provided in the embodiment of the present application;
FIG. 5 is a flowchart of an event feature extraction method according to an embodiment of the present disclosure;
fig. 6 is a schematic diagram of the substeps of S104 provided in the embodiment of the present application;
fig. 7 is a schematic unit diagram of an event feature extraction device according to an embodiment of the present application.
In the figure: 10-a processor; 11-memory; 12-bus; 13-a communication interface; 201-a processing unit; 202-fusion unit.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. The components of the embodiments of the present application, which are generally described and illustrated in the figures herein, may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present application, as provided in the accompanying drawings, is not intended to limit the scope of the application, as claimed, but is merely representative of selected embodiments of the application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.
It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures. Meanwhile, in the description of the present application, the terms "first", "second", and the like are used only to distinguish the description, and are not to be construed as indicating or implying relative importance.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
In the description of the present application, it should be noted that, the terms "upper," "lower," "inner," "outer," and the like indicate an orientation or a positional relationship based on the orientation or the positional relationship shown in the drawings, or an orientation or a positional relationship conventionally put in use of the product of the application, merely for convenience of description and simplification of the description, and do not indicate or imply that the apparatus or element to be referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present application.
In the description of the present application, it should also be noted that, unless explicitly specified and limited otherwise, the terms "disposed," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the terms in this application will be understood by those of ordinary skill in the art in a specific context.
Some embodiments of the present application are described in detail below with reference to the accompanying drawings. The following embodiments and features of the embodiments may be combined with each other without conflict.
An event refers to a change in state that occurs at a particular point in time or time period, within a particular geographic area, and that consists of one or more actions that one or more characters are engaged in. With the development of network science and technology, massive text information can be derived from an event in the process of generating and spreading, and how to propose event features from the massive text information is a key step of event analysis. Existing event feature extraction methods often only focus on internal features of an event, such as people, places, time, etc. in the event. Neglecting that events derive massive text information during the generation and propagation process with the development of network science and technology, external features among a large number of events may be derived. This is also a technical problem to be solved by the person skilled in the art.
The embodiment of the application provides electronic equipment, which can be server equipment. Referring to fig. 1, a schematic structure of an electronic device is shown. The electronic device comprises a processor 10, a memory 11, a bus 12. The processor 10 and the memory 11 are connected by a bus 12, the processor 10 being adapted to execute executable modules, such as computer programs, stored in the memory 11.
The processor 10 may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the event feature extraction method may be performed by integrated logic circuitry of hardware in the processor 10 or by instructions in the form of software. The processor 10 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), etc.; but also digital signal processors (Digital Signal Processor, DSP for short), application specific integrated circuits (Application Specific Integrated Circuit, ASIC for short), field-programmable gate arrays (Field-Programmable Gate Array, FPGA for short) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
The memory 11 may comprise a high-speed random access memory (RAM: random Access Memory) and may also comprise a non-volatile memory (non-volatile memory), such as at least one disk memory.
Bus 12 may be a ISA (Industry Standard Architecture) bus, PCI (Peripheral Component Interconnect) bus, EISA (Extended Industry Standard Architecture) bus, or the like. Only one double-headed arrow is shown in fig. 1, but not only one bus 12 or one type of bus 12.
The memory 11 is used for storing programs such as programs corresponding to the event feature extraction means. The event feature extraction means comprise at least one software function module which may be stored in the memory 11 in the form of software or firmware (firmware) or cured in the Operating System (OS) of the electronic device. The processor 10, upon receiving the execution instruction, executes the program to implement the event feature extraction method.
Possibly, the electronic device provided in the embodiment of the present application further includes a communication interface 13. The communication interface 13 is connected to the processor 10 via a bus. The electronic device may communicate with other terminals via the communication interface 13 to receive data transmitted by the other terminals.
It should be understood that the structure shown in fig. 1 is a schematic structural diagram of only a portion of an electronic device, which may also include more or fewer components than shown in fig. 1, or have a different configuration than shown in fig. 1. The components shown in fig. 1 may be implemented in hardware, software, or a combination thereof.
The method for extracting event features provided by the embodiment of the invention can be applied to, but is not limited to, the electronic device shown in fig. 1, and the specific flow is shown in fig. 2:
s105, taking the event text as input of a node neural network model, and outputting corresponding internal characteristics by the node neural network model.
The internal features are feature representations generated by combining all internal nodes in the event, and the internal nodes are words in the event text.
For example, the event text is "Mister A and Miss B in D civil office collar certificate on day C", and Mister A, mister B, on day C, in, D civil office and collar certificate are words in the event text, and are interior nodes. The internal features generated by the different internal nodes combined together are not identical.
S106, taking the event theme as input of an associated neural network model, and outputting corresponding external characteristics by the associated neural network model according to the relation network.
Wherein, the event theme corresponds to the event text; the relationship network contains events and other events having direct or indirect relationship with the events, and the external features are feature representations of the degree of association of the events with other events.
Specifically, the direct or indirect relationship includes a co-transmission relationship in which two events are transmitted by the same carrier, a co-domain relationship in which two events occur identically, a co-evaluation relationship in which two events have identical comments or commentators, a simultaneous relationship in which two events occur identically, and the like. The nature and specific form of the direct or indirect relationship are not limited, but of course, the direct or indirect relationship may also include other relationships.
The association of different events with other times in the relationship network may not be exactly the same, and correspondingly, external features corresponding to different events may be different.
S107, taking the internal features and the external features as the input of a fusion network model, and outputting event features corresponding to the events by the fusion network model.
Specifically, the fusion network model fuses the input internal features and external features to obtain corresponding event features. The event features cover both the combined feature representation of the internal nodes of the event text and the feature representation of the association relationship between the event and other events, namely the feature representation of the event features as the whole of the event. In contrast, in the conventional event feature extraction method, only the internal features of the event, such as the person, place, time, etc. in the event are often focused on. According to the event feature extraction method provided by the embodiment of the application, the fact that massive text information can be derived in the process of generating and spreading the events along with the development of network science and technology is considered, and external features among a large number of events can be derived. And the external characteristics and the internal characteristics are fused, so that the obtained event characteristics are more perfect, and the subsequent data processing is more convenient.
In summary, in the event feature extraction method provided in the embodiment of the present application, first, an event text is used as an input of a node neural network model, the node neural network model outputs a corresponding internal feature, an event topic is used as an input of an associated neural network model, and the associated neural network model outputs a corresponding external feature according to a relationship network; taking the internal features and the external features as inputs of a fusion network model, and outputting event features corresponding to the events by the fusion network model; the internal characteristics of the event are concerned, and the fact that massive text information and external characteristics among a large number of events can be derived in the process of generating and transmitting the event is considered as the network science and technology develop. And the external characteristics and the internal characteristics are fused, so that the obtained event characteristics are more perfect, and the subsequent data processing is more convenient.
On the basis of fig. 2, for the content in S105, the embodiment of the present application further provides a possible implementation manner, please refer to fig. 3, S105 includes:
s105-1, taking the event text as input of the node neural network model.
Possibly, the event text is loaded into the node neural network model as its input.
S105-2, constructing a syntactic network diagram according to the syntactic relation.
The syntactic network diagram comprises all internal nodes and connection relations among the internal nodes.
Possibly, a syntactic parse tree is built from syntactic dependencies, such as main-predicate relationships, guest-move relationships, etc., using a syntactic analysis tool based on the descriptive statements in the event text. And then converted into a corresponding syntactic network map.
G′＝[G′ 1 ,G′ 2 ,…,G′ n ]
G′ j ＝(V,E)
Wherein G 'characterizes a syntactic network map, G' j And characterizing a jth event syntax network, wherein V represents nodes of the jth event syntax network, and E represents the relation between the nodes in the jth event syntax network.
S105-3, constructing a first adjacency matrix according to the syntactic network diagram.
The first adjacency matrix characterizes whether any one internal node and other internal nodes have a connection relation or not. The connection between an internal node and other internal nodes may be syntactic dependencies.
And constructing an adjacency matrix corresponding to the syntactic network based on the constructed event syntactic network diagram. The adjacency matrix is as follows:
wherein A 'is' j Representing an adjacency matrix corresponding to a jth event syntax network, wherein L represents the number of nodes in the jth event syntax network, and a pq Representing syntactic dependencies between nodes p and q in a syntactic network, a pq =0 indicates that there is no syntactic dependency between the two, a pq =1 indicates that there is a syntactic dependency between the two.
S105-4, performing aggregation processing on the characteristic representation of the internal node according to the first adjacent matrix to obtain an internal characteristic, and inputting the internal characteristic.
Specifically, feature representations of internal nodes having a connection relationship (syntactic dependency relationship) are subjected to aggregation processing to obtain internal features, the formula is as follows:
M j ＝softmax(X′ j W Q (X′ j W K ) T )
wherein X 'is' j Representing node characteristics in a jth event syntax network, W Q And W is K Network model parameters, T is transposed symbol, M j Similarity matrix formed for all internal nodes, A' j Representing adjacency matrix corresponding to the jth event syntax network, I' j Representing A' j Corresponding identity matrix, W 3 、W 4 、b 3 、b 4 For network model parameters, σ is the activation function, f' j Representing internal features.
Possibly, based on the pre-trained chinese word vector, a feature representation of an interior node in the syntactic network is obtained or a feature representation of each interior node in the network graph is randomly initialized.
On the basis of fig. 2, for the content in S106, the embodiment of the present application further provides a possible implementation manner, please refer to fig. 4, S106 includes:
s106-1, taking the event theme as input of the associated neural network model.
Possibly, the event topic may be constituted by important nodes that extract the event text, including time, people, places and actions.
S106-2, constructing an event relation network graph based on the external nodes shared by the event and any other events.
The relationship network graph characterizes the common relationship among the events, and the external nodes are nodes which are outside the text content of the events and are associated with the events.
Specifically, based on the constructed external event network, a relationship network tree between events is constructed according to external nodes shared between different events, for example, if two events have the same place of occurrence, the two events are in the same domain relationship, the two events have the same forwarding person, and the two events are in the same transmission relationship.
Based on the constructed event relation network tree, layering the relation network according to different relations to obtain an event relation network diagram.
G＝[G 1 ,G 2 ,…,G M ]
G l ＝(V,E l )
Wherein G represents an event relationship network diagram, G l Representing a J-th event relation network, M representing the number of event relation networks, V representing event nodes in the J-th event relation network, the number of the nodes being N, the number of the N being the same, E l Representing the connection edge of the relation network corresponding to the first external node.
S106-3, constructing a second adjacency matrix according to the event relation network diagram.
Wherein the second adjacency matrix characterizes whether any event has a common relationship with other events.
The second adjacency matrix corresponding to the event relationship network graph is as follows:
wherein A is l Representing an adjacency matrix corresponding to the first event relation network, N represents the number of event nodes in the event relation network, and a mn Representing the connection relationship between nodes m and n in a network, a mn =0 indicates that there is no connection between the two, a mn =1 indicates that there is a connection relationship between the two.
S106-4, performing aggregation processing on the characteristic representations of the external nodes according to the second adjacent matrix to generate external characteristics, and outputting the external characteristics.
Based on the constructed second adjacency matrix, aggregating the feature representation of the external nodes of the event to obtain the external features of the event, wherein the main expression is as follows:
wherein A is l Representing the adjacency matrix corresponding to the first event relationship network, I l Representation A l Corresponding identity matrix, the dimension is the same as the dimension of corresponding adjacent matrix, D l Representation matrix A l Degree value matrix of W l And b l X is a network model parameter l Feature input representing the first event relationship network, w' l The weight coefficients representing different relation networks are obtained by training, and F represents external characteristics.
On the basis of fig. 2, for the content in S107, the embodiment of the present application further provides a possible implementation, please refer to the following.
Based on the obtained internal and external features, an event feature representation is obtained:
f j ″＝αf′ j +βf j
wherein f j "event feature representing jth event, f' j Representing internal features of the jth event, f j E F represents the external feature of the j-th event, and α and β are hyper-parameters representing the weight coefficients of the internal and external features, respectively.
On the basis of fig. 2, regarding how to train the neural network model, a possible implementation manner is further provided in the embodiments of the present application, referring to fig. 5, the event feature extraction method further includes:
and S101, training the node neural network model, the associated neural network model and the fusion network model according to the training event text carrying the label to obtain training event characteristics corresponding to the training event.
Wherein the tag is the actual type of event.
S102, performing dimension reduction processing on the training event features through a first type of nonlinear transformation function to obtain low-dimension training event features.
Based on the training event characteristics, the low-dimensional training event characteristics corresponding to the training events are obtained through a first nonlinear transformation function, and the main formulas are as follows:
wherein,representing low-dimensional training event features corresponding to the jth training event, relu represents the activation function, f j "represent training event features corresponding to the jth training event, W 5 、b 5 Representing network model parameters.
S103, processing the low-dimensional training event features through a second type nonlinear transformation function to obtain a prediction type corresponding to the training event.
Based on the low-dimensional training event characteristics and the second nonlinear transformation function, outputting a prediction type corresponding to the training event, wherein the main formula is as follows:
wherein,representing the predicted type of the jth training event, sigma represents the activation function, W 6 、b 6 Representing model parameters->Representing the low-dimensional training event characteristics of the jth training event.
And S104, optimizing the node neural network model, the associated neural network model and the fusion network model according to the labels and the prediction types.
Specifically, the label is the real type of the training event, and the node neural network model, the associated neural network model and the fusion network model can be optimized according to the comparison result of the label and the prediction type, namely the recognition accuracy.
On the basis of fig. 5, for the content in S104, a possible implementation manner is further provided in the embodiment of the present application, please refer to fig. 6, and S104 includes:
s104-1, constructing a loss function according to the label and the prediction type.
The expression of the loss function is:
wherein,representing the predicted type of the jth training event, L representing the model loss function, N representing the total number of events, y j Representing the true type of the jth training event.
S104-2, stopping training when the loss function meets the preset condition.
Specifically, when the preset condition may be that the training number reaches the preset number, or the loss function is smaller than the preset threshold, or the loss function reaches the locally optimal solution.
Referring to fig. 7, an event feature extraction apparatus according to an embodiment of the present application is provided, and optionally, the event feature extraction apparatus is applied to the electronic device described above.
The event feature extraction device includes: a processing unit 201 and a fusion unit 202.
The processing unit 201 is configured to take the event text as input of a node neural network model, where the node neural network model outputs corresponding internal features, where the internal features are feature representations generated by combining internal nodes in the event, and the internal nodes are words in the event text; the method is also used for taking the event theme as input of an associated neural network model, and outputting corresponding external features by the associated neural network model according to a relational network, wherein the event theme corresponds to an event text; the relationship network contains events and other events having direct or indirect relationship with the events, and the external features are feature representations of the degree of association of the events with other events. Specifically, the processing unit 201 may execute S105 and S106 described above.
The fusion unit 202 is configured to take the internal feature and the external feature as inputs of a fusion network model, and output an event feature corresponding to the event by the fusion network model, where the event feature is a feature representation of the whole event. Specifically, the fusion unit 202 may perform S107 described above.
Further, the processing unit 201 is configured to take the event text as an input of the node neural network model; constructing a syntactic network diagram according to the syntactic relation, wherein the syntactic network diagram comprises all internal nodes and connection relations among the internal nodes; constructing a first adjacency matrix according to the syntactic network diagram, wherein the first adjacency matrix represents whether any one internal node and other internal nodes have a connection relation or not; and carrying out aggregation processing on the characteristic representations of the internal nodes according to the first adjacent matrix to obtain internal characteristics, and outputting the internal characteristics. Specifically, the processing unit 201 may execute S105-1 to S105-4 described above.
Further, the processing unit 201 is configured to take the event topic as an input of the associated neural network model; constructing an event relationship network diagram based on external nodes shared by the events and any other events, wherein the relationship network diagram characterizes the shared relationship among the events, and the external nodes are nodes which are outside the text content of the events and are associated with the events; constructing a second adjacency matrix according to the event relation network diagram, wherein the second adjacency matrix characterizes whether any event and other events have a common relation or not; and carrying out aggregation processing on the characteristic representations of the external nodes according to the second adjacency matrix to generate external characteristics, and outputting the external characteristics. Specifically, the processing unit 201 may execute the above-described S106-1 to S106-4.
It should be noted that, the event feature extraction apparatus provided in this embodiment may execute the method flow shown in the method flow embodiment to achieve the corresponding technical effects. For a brief description, reference is made to the corresponding parts of the above embodiments, where this embodiment is not mentioned.
The embodiment of the invention also provides a storage medium storing computer instructions and programs which when read and executed perform the event feature extraction method of the embodiment. The storage medium may include memory, flash memory, registers, combinations thereof, or the like.
An electronic device, which may be a server device, is provided below, where the electronic device is shown in fig. 1, and the event feature extraction method described above may be implemented; specifically, the electronic device includes: a processor 10, a memory 11, a bus 12. The processor 10 may be a CPU. The memory 11 is used to store one or more programs that, when executed by the processor 10, perform the event feature extraction method of the above-described embodiments.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners as well. The apparatus embodiments described above are merely illustrative, for example, flow diagrams and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present application may be integrated together to form a single part, or each module may exist alone, or two or more modules may be integrated to form a single part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The foregoing description is only of the preferred embodiments of the present application and is not intended to limit the same, but rather, various modifications and variations may be made by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.
It will be evident to those skilled in the art that the present application is not limited to the details of the foregoing illustrative embodiments, and that the present application may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (10)
1. A method of event feature extraction, the method comprising:
taking an event text as input of a node neural network model, and outputting corresponding internal features by the node neural network model, wherein the internal features are feature representations generated by combining all internal nodes in an event, and the internal nodes are words in the event text;
taking an event theme as input of an associated neural network model, and outputting corresponding external features by the associated neural network model according to a relational network, wherein the event theme corresponds to the event text, and the event theme is formed by extracting important nodes of the event text, wherein the important nodes comprise time, characters, places and actions; the relation network comprises the event and other events which have direct or indirect relation with the event, and the external characteristics are characteristic representations of the association degree of the event and the other events;
and taking the internal features and the external features as inputs of a fusion network model, and outputting event features corresponding to the event by the fusion network model, wherein the event features are integral feature representations of the event.
2. The event feature extraction method as claimed in claim 1, wherein the step of using the event text as an input to a node neural network model, and outputting the corresponding internal feature from the node neural network model, comprises:
taking the event text as input of a node neural network model;
constructing a syntactic network diagram according to the syntactic relation, wherein the syntactic network diagram comprises all internal nodes and connection relations among the internal nodes;
constructing a first adjacency matrix according to the syntactic network diagram, wherein the first adjacency matrix represents whether any one internal node and other internal nodes have a connection relationship or not;
and carrying out aggregation processing on the characteristic representation of the internal node according to the first adjacency matrix to obtain the internal characteristic, and outputting the internal characteristic.
3. The event feature extraction method as claimed in claim 1, wherein the step of using the event topic as an input of an associated neural network model, the associated neural network model outputting the corresponding external feature according to a relational network, comprises:
taking the event theme as input of the associated neural network model;
constructing an event relationship network graph based on external nodes shared by the event and any other events, wherein the relationship network graph represents the shared relationship among the events, and the external nodes are nodes which are outside the text content of the event and are associated with the event;
constructing a second adjacency matrix according to the event relation network diagram, wherein the second adjacency matrix represents whether any event and other events have a common relation or not;
and carrying out aggregation processing on the characteristic representation of the external node according to the second adjacency matrix to generate the external characteristic, and outputting the external characteristic.
4. The event feature extraction method of claim 1, wherein the method further comprises:
training the node neural network model, the associated neural network model and the fusion network model according to training event texts carrying labels to obtain training event characteristics corresponding to training events, wherein the labels are real types of the events;
performing dimension reduction processing on the training event features through a first type nonlinear transformation function to obtain low-dimension training event features;
processing the low-dimensional training event characteristics through a second class nonlinear transformation function to obtain a prediction type corresponding to the training event;
and optimizing the node neural network model, the associated neural network model and the fusion network model according to the labels and the prediction types.
5. The event feature extraction method of claim 4, wherein the optimizing the node neural network model, the associated neural network model, and the fusion network model according to the label and the prediction type comprises:
constructing a loss function according to the label and the prediction type;
the expression of the loss function is:
wherein,representing the predicted type of the jth training event, L representing the model loss function, N representing the total number of events, y j Representing the true category of the jth training event;
and stopping training when the loss function meets a preset condition.
6. An event feature extraction apparatus, the apparatus comprising:
the processing unit is used for taking the event text as the input of a node neural network model, and outputting corresponding internal features by the node neural network model, wherein the internal features are feature representations generated by combining all internal nodes in the event, and the internal nodes are words in the event text; the method is also used for inputting an event theme as an associated neural network model, and the associated neural network model outputs corresponding external features according to a relational network, wherein the event theme corresponds to the event text, and the event theme is formed by extracting important nodes of the event text, wherein the important nodes comprise time, characters, places and actions; the relation network comprises the event and other events which have direct or indirect relation with the event, and the external characteristics are characteristic representations of the association degree of the event and the other events;
and the fusion unit is used for taking the internal features and the external features as the input of a fusion network model, and outputting event features corresponding to the events by the fusion network model, wherein the event features are integral feature representations of the events.
7. The event feature extraction apparatus of claim 6, wherein the processing unit is configured to take event text as input to a node neural network model; constructing a syntactic network diagram according to the syntactic relation, wherein the syntactic network diagram comprises all internal nodes and connection relations among the internal nodes; constructing a first adjacency matrix according to the syntactic network diagram, wherein the first adjacency matrix represents whether any one internal node and other internal nodes have a connection relationship or not; and carrying out aggregation processing on the characteristic representation of the internal node according to the first adjacency matrix to obtain the internal characteristic, and outputting the internal characteristic.
8. The event feature extraction apparatus of claim 6 wherein the processing unit is configured to take an event topic as input to an associated neural network model; constructing an event relationship network graph based on external nodes shared by the event and any other events, wherein the relationship network graph represents the shared relationship among the events, and the external nodes are nodes which are outside the text content of the event and are associated with the event; constructing a second adjacency matrix according to the event relation network diagram, wherein the second adjacency matrix represents whether any event and other events have a common relation or not; and carrying out aggregation processing on the characteristic representation of the external node according to the second adjacency matrix to generate the external characteristic, and outputting the external characteristic.
9. A storage medium having stored thereon a computer program which, when executed by a processor, implements the method of any of claims 1-5.
10. An electronic device, comprising: a processor and a memory for storing one or more programs; the method of any of claims 1-5 is implemented when the one or more programs are executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010946188.4A CN112085104B (en) | 2020-09-10 | 2020-09-10 | Event feature extraction method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010946188.4A CN112085104B (en) | 2020-09-10 | 2020-09-10 | Event feature extraction method and device, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112085104A CN112085104A (en) | 2020-12-15 |
| CN112085104B true CN112085104B (en) | 2024-04-12 |
Family
ID=73732240
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010946188.4A Active CN112085104B (en) | 2020-09-10 | 2020-09-10 | Event feature extraction method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112085104B (en) |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102929918A (en) * | 2012-09-20 | 2013-02-13 | 西北工业大学 | False online public opinion identification method |
| CN104899657A (en) * | 2015-06-09 | 2015-09-09 | 北京邮电大学 | Method for predicting association fusion events |
| CN109726289A (en) * | 2018-12-29 | 2019-05-07 | 北京百度网讯科技有限公司 | Event detection method and device |
| CN110377738A (en) * | 2019-07-15 | 2019-10-25 | 昆明理工大学 | Merge the Vietnamese news event detecting method of interdependent syntactic information and convolutional neural networks |
| CN110457757A (en) * | 2019-07-16 | 2019-11-15 | 江西理工大学 | Prediction method and device of rock mass instability stage based on multi-feature fusion |
| CN110458466A (en) * | 2019-08-16 | 2019-11-15 | 内蒙古大学 | Patent Valuation Method and Valuation System Based on Data Mining and Heterogeneous Knowledge Association |
| CN110516077A (en) * | 2019-08-20 | 2019-11-29 | 北京中亦安图科技股份有限公司 | Knowledge mapping construction method and device towards enterprise's market conditions |
| WO2020001373A1 (en) * | 2018-06-26 | 2020-01-02 | 杭州海康威视数字技术股份有限公司 | Method and apparatus for ontology construction |
| CN111178039A (en) * | 2019-12-18 | 2020-05-19 | 北京明略软件系统有限公司 | Model training method and device, and method and device for realizing text processing |
| CN111260261A (en) * | 2020-03-12 | 2020-06-09 | 广州宝顺信息科技有限公司 | System probability risk automatic evaluation method |
| WO2020114122A1 (en) * | 2018-12-07 | 2020-06-11 | 阿里巴巴集团控股有限公司 | Neural network system and method for analyzing relationship network graph |
| CN111367961A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Time series data event prediction method, system and application based on graph convolutional neural network |
| CN111488726A (en) * | 2020-03-31 | 2020-08-04 | 成都数之联科技有限公司 | Pointer network-based unstructured text extraction multi-task joint training method |
| CN111488740A (en) * | 2020-03-27 | 2020-08-04 | 北京百度网讯科技有限公司 | Causal relationship judging method and device, electronic equipment and storage medium |
| US10757061B1 (en) * | 2016-08-18 | 2020-08-25 | Hrl Laboratories, Llc | Cross-modal event summarization system based on network of networks representations |
| CN111581983A (en) * | 2020-05-13 | 2020-08-25 | 中国人民解放军国防科技大学 | Method for predicting social concern hotspots in network public opinion events based on group analysis |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3291120B1 (en) * | 2016-09-06 | 2021-04-21 | Accenture Global Solutions Limited | Graph database analysis for network anomaly detection systems |
| US10572601B2 (en) * | 2017-07-28 | 2020-02-25 | International Business Machines Corporation | Unsupervised template extraction |
-
2020
- 2020-09-10 CN CN202010946188.4A patent/CN112085104B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102929918A (en) * | 2012-09-20 | 2013-02-13 | 西北工业大学 | False online public opinion identification method |
| CN104899657A (en) * | 2015-06-09 | 2015-09-09 | 北京邮电大学 | Method for predicting association fusion events |
| US10757061B1 (en) * | 2016-08-18 | 2020-08-25 | Hrl Laboratories, Llc | Cross-modal event summarization system based on network of networks representations |
| WO2020001373A1 (en) * | 2018-06-26 | 2020-01-02 | 杭州海康威视数字技术股份有限公司 | Method and apparatus for ontology construction |
| WO2020114122A1 (en) * | 2018-12-07 | 2020-06-11 | 阿里巴巴集团控股有限公司 | Neural network system and method for analyzing relationship network graph |
| CN109726289A (en) * | 2018-12-29 | 2019-05-07 | 北京百度网讯科技有限公司 | Event detection method and device |
| CN110377738A (en) * | 2019-07-15 | 2019-10-25 | 昆明理工大学 | Merge the Vietnamese news event detecting method of interdependent syntactic information and convolutional neural networks |
| CN110457757A (en) * | 2019-07-16 | 2019-11-15 | 江西理工大学 | Prediction method and device of rock mass instability stage based on multi-feature fusion |
| CN110458466A (en) * | 2019-08-16 | 2019-11-15 | 内蒙古大学 | Patent Valuation Method and Valuation System Based on Data Mining and Heterogeneous Knowledge Association |
| CN110516077A (en) * | 2019-08-20 | 2019-11-29 | 北京中亦安图科技股份有限公司 | Knowledge mapping construction method and device towards enterprise's market conditions |
| CN111178039A (en) * | 2019-12-18 | 2020-05-19 | 北京明略软件系统有限公司 | Model training method and device, and method and device for realizing text processing |
| CN111367961A (en) * | 2020-02-27 | 2020-07-03 | 西安交通大学 | Time series data event prediction method, system and application based on graph convolutional neural network |
| CN111260261A (en) * | 2020-03-12 | 2020-06-09 | 广州宝顺信息科技有限公司 | System probability risk automatic evaluation method |
| CN111488740A (en) * | 2020-03-27 | 2020-08-04 | 北京百度网讯科技有限公司 | Causal relationship judging method and device, electronic equipment and storage medium |
| CN111488726A (en) * | 2020-03-31 | 2020-08-04 | 成都数之联科技有限公司 | Pointer network-based unstructured text extraction multi-task joint training method |
| CN111581983A (en) * | 2020-05-13 | 2020-08-25 | 中国人民解放军国防科技大学 | Method for predicting social concern hotspots in network public opinion events based on group analysis |
Non-Patent Citations (2)
| Title |
|---|
| 基于复杂网络的食品安全事件新闻文本可视化及分析;江美辉;安海忠;高湘昀;管青;郝晓晴;;情报杂志(12);全文 * |
| 融合依存关系和篇章修辞关系的事件时序关系识别;戴倩雯;张龙印;孔芳;;模式识别与人工智能(12);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112085104A (en) | 2020-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Staworko et al. | Complexity and Expressiveness of ShEx for RDF | |
| CN110909160A (en) | Regular expression generation method, server and computer-readable storage medium | |
| WO2019119967A1 (en) | Method and device using wikipedia link structure to generate chinese language concept vector | |
| CN111194401B (en) | Abstraction and portability of intent recognition | |
| CN110674188A (en) | Feature extraction method, device and equipment | |
| CN109948160B (en) | Short text classification method and device | |
| KR20210063874A (en) | A method and an apparatus for analyzing marketing information based on knowledge graphs | |
| CN113360654A (en) | Text classification method and device, electronic equipment and readable storage medium | |
| Zhang et al. | User classification with multiple textual perspectives | |
| CN112232070A (en) | Natural language processing model construction method, system, electronic device and storage medium | |
| CN114547310A (en) | A method, system, device and medium for early detection of fake news | |
| WO2025000910A1 (en) | Keyword category identification method and apparatus, and electronic device | |
| Li et al. | Nlp-adbench: Nlp anomaly detection benchmark | |
| CN112085104B (en) | Event feature extraction method and device, storage medium and electronic equipment | |
| US20250103722A1 (en) | Hierarchical representation models | |
| CN117436457B (en) | Irony identification method, irony identification device, computing equipment and storage medium | |
| CN112069822A (en) | Method, device and equipment for acquiring word vector representation and readable medium | |
| CN118364117A (en) | Government affair question-answering method based on knowledge graph and related equipment | |
| CN116702895A (en) | Domain-extensible knowledge graph construction method, device, equipment and medium | |
| Nanduri et al. | Modified Fuzzy Approach to Automatic Classification of Cyber Hate Speech from the Online Social Networks (OSN's). | |
| De Oliveira et al. | Advances in the study of elementary cellular automata regular language complexity | |
| CN112364131B (en) | Corpus processing method and related device thereof | |
| CN114564958A (en) | Text recognition method, device, equipment and medium | |
| CN113037715B (en) | Transmission system and method for limiting manipulation of content in a network environment | |
| CN111881928B (en) | Coding model training method and device, storage medium and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |