CN111241209B - Method and apparatus for generating information - Google Patents

Abstract

The embodiment of the application discloses a method and a device for generating information, and belongs to the fields of computer technology and knowledge maps. One embodiment of the method comprises the following steps: acquiring a main body relation binary group and a text, wherein the main body relation binary group comprises a main body and a relation; segmenting the text into text word sequences; inputting the subject relation binary groups and the text word sequences into a pre-trained groove filling model to obtain a labeling result of the text word sequences, wherein the groove filling model is used for labeling objects in the text word sequences; and generating a subject relationship object triplet based on the subject relationship triplet and the labeling result, wherein the subject relationship object triplet comprises a subject, a relationship and an object of the text. This embodiment improves object recognition accuracy.

Description

Method and apparatus for generating information

technical field

The embodiments of the present application relate to the field of computer technologies, and in particular to methods and devices for generating information.

Background technique

A knowledge graph is a large-scale knowledge base of real-world knowledge expressed in a structured form from a semantic point of view. It is a directed graph, which includes elements such as entities (nodes) and relationships (edges). SPO (Subject Predication Object, subject predicate object) triples are also called subject-relationship-object triples, which refer to triples composed of entity pairs (S and O) and their relationship (P). From the perspective of knowledge graph construction, entity relationship extraction can obtain the relationship attribute values of entities missing, which is used to improve the connectivity of knowledge graphs, and effectively improve the knowledge richness and completeness of knowledge graphs.

At present, the commonly used entity extraction method is to convert the subject-relationship pair into a question, and input the question and text into the reading comprehension model, and the reading comprehension model will mark the starting and ending positions of the object in the text. However, the reading comprehension model actually degenerates the subject-relationship dyad into a problem, which loses the structural information and affects the object recognition effect.

Contents of the invention

The embodiments of the present application propose a method and an apparatus for generating information.

In the first aspect, the embodiment of the present application proposes a method for generating information to obtain subject-relationship dyads and texts, wherein the subject-relationship dyads include subjects and relations; texts are segmented into text word sequences; subjects The relational dyads and text word sequences are input to the pre-trained slot filling model to obtain the labeling results of the text word sequences, wherein the slot filling model is used to label the objects in the text word sequences; based on the subject-relational dyads and the labeling results, A subject-relation-object triplet is generated, wherein the subject-relation-object triplet includes the subject, relation and object of the text.

In some embodiments, the slot-filling model includes an input layer, a localization layer, an embedding layer, an encoding layer, a decoding layer, and an output layer.

In some embodiments, the subject-relationship pair and the text word sequence are input to the pre-trained slot filling model to obtain the labeling result of the text word sequence, including: input the subject-relationship pair and the text word sequence into the input layer, Obtain the word sequence feature and distance feature; input the distance feature to the positioning layer to obtain position information; input the word sequence feature and position information to the embedding layer to obtain the word sequence vector and position vector; input the word sequence vector to the encoding layer to obtain Encoding vector; input the position vector and encoding vector to the decoding layer to obtain the decoding vector; input the decoding vector to the output layer to obtain the labeling result.

In some embodiments, the encoding layer includes a first bidirectional LSTM network, and the decoding layer includes a positional attention module, a relational attention module, and a second bidirectional LSTM network.

In some embodiments, the position vector and the encoding vector are input to the decoding layer to obtain the decoding vector, including: inputting the splicing of the position vector and the encoding vector to the position attention module to obtain the word distance subject and relation in the text word sequence Position information; input the long-short-term memory network encoding and encoding vector of the relationship into the relationship attention module to obtain the semantic similarity between the words in the text word sequence and the relationship; the encoding vector, the word distance in the text word sequence and the relationship The location information and the semantic similarity between words and relations in the text word sequence are input to the second bidirectional long-short-term memory network to obtain a decoding vector.

In some embodiments, the decoding vector is input to the output layer to obtain the labeling result, including: performing multi-classification on the decoding vector of the words in the text word sequence through the activation function, and obtaining the words in the text word sequence belonging to multiple categories The probability of each category, wherein, the multi-classification is to calculate the probability of words belonging to each category in multiple categories; the text word sequence is marked based on the category corresponding to the maximum probability of the word in the text word sequence, and the labeling result is generated.

In some embodiments, the word sequence feature includes at least one of the following: a text word sequence, a part-of-speech sequence of a text word sequence, a named entity recognition sequence of a text word sequence, and a relational word sequence of a relationship, and the distance feature includes at least one of the following: text Distances from words in word sequences to subjects, and distances from words in text word sequences to relations.

In some embodiments, the slot-filling model uses a BIOES sequence tagging method to tag text word sequences.

In the second aspect, the embodiment of the present application proposes a device for generating information, including: an acquisition unit configured to acquire a subject-relationship tuple and text, wherein the subject-relationship tuple includes subject and relationship; segmentation The unit is configured to divide the text into text word sequences; the labeling unit is configured to input the subject-relationship binary and the text word sequence into the pre-trained slot filling model to obtain the labeling result of the text word sequence, wherein the slot filling The model is used to mark the object in the text word sequence; the generating unit is configured to generate a subject-relation-object triplet based on the subject-relationship pair and the tagging result, wherein the subject-relationship-object triplet includes the subject, relation and object.

In some embodiments, the slot-filling model includes an input layer, a localization layer, an embedding layer, an encoding layer, a decoding layer, and an output layer.

In some embodiments, the labeling unit includes: an input subunit configured to input the subject-relationship pair and the text word sequence to the input layer to obtain the word sequence feature and the distance feature; the localization subunit is configured to input the distance feature Input to the positioning layer to obtain position information; the embedding subunit is configured to input the word sequence feature and position information to the embedding layer to obtain the word sequence vector and position vector; the encoding subunit is configured to input the word sequence vector to the encoding layer to obtain the encoding vector; the decoding subunit is configured to input the position vector and the encoding vector to the decoding layer to obtain the decoding vector; the output subunit is configured to input the decoding vector to the output layer to obtain the labeling result.

In some embodiments, the encoding layer includes a first bidirectional LSTM network, and the decoding layer includes a positional attention module, a relational attention module, and a second bidirectional LSTM network.

In some embodiments, the encoding subunit is further configured to: input the concatenation of the position vector and the encoding vector to the position attention module to obtain the position information of the distance between the word in the text word sequence and the relationship between the subject and the relationship; The network encoding and encoding vector are input to the relationship attention module to obtain the semantic similarity between the words in the text word sequence and the relationship; The semantic similarity with the relationship is input to the second bidirectional long-short-term memory network to obtain a decoding vector.

In some embodiments, the output subunit is further configured to: perform multi-classification on the decoding vectors of the words in the text word sequence through an activation function to obtain the probability that the words in the text word sequence belong to each category in multiple categories, Among them, multi-classification is to calculate the probability that words belong to each category in multiple categories; based on the category corresponding to the maximum probability of words in the text word sequence, the text word sequence is tagged to generate tagging results.

In some embodiments, the word sequence feature includes at least one of the following: a text word sequence, a part-of-speech sequence of a text word sequence, a named entity recognition sequence of a text word sequence, and a relational word sequence of a relationship, and the distance feature includes at least one of the following: text Distances from words in word sequences to subjects, and distances from words in text word sequences to relations.

In some embodiments, the slot-filling model uses a BIOES sequence tagging method to tag text word sequences.

In the third aspect, the embodiment of the present application provides an electronic device, the electronic device includes: one or more processors; a storage device, on which one or more programs are stored; when one or more programs are used by one or more processors, so that one or more processors implement the method described in any implementation manner in the first aspect.

In a fourth aspect, an embodiment of the present application provides a computer-readable medium, on which a computer program is stored, and when the computer program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

The method and device for generating information provided by the embodiments of the present application firstly obtain subject-relationship dyads and texts; then text is segmented into text word sequences; then subject-relationship dyads and text word sequences are input to the pre-trained The slot-filling model is used to obtain the tagging results of the text word sequence; finally, based on the subject-relationship doublets and the tagging results, the subject-relationship-object triplets are generated. The object is recognized based on the slot-filling model, which retains the sum structure of the subject and the relationship, and improves the accuracy of object recognition.

Description of drawings

Other characteristics, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

FIG. 1 is an exemplary system architecture to which the present application can be applied;

Figure 2 is a flowchart of one embodiment of a method for generating information according to the present application;

FIG. 3 is a flowchart of another embodiment of a method for generating information according to the present application;

Fig. 4 shows the schematic diagram of the structure of the slot filling model;

Fig. 5 is a schematic structural diagram of an embodiment of a device for generating information according to the present application;

Fig. 6 is a schematic structural diagram of a computer system suitable for implementing the electronic device of the embodiment of the present application.

Detailed ways

The application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain related inventions, rather than to limit the invention. It should also be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.

It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and embodiments.

FIG. 1 shows an exemplary system architecture 100 to which embodiments of the method for generating information or the apparatus for generating information of the present application can be applied.

As shown in FIG. 1 , a system architecture 100 may include a terminal device 101 , a network 102 and a server 103 . The network 102 is used as a medium for providing a communication link between the terminal device 101 and the server 103 . Network 102 may include various connection types, such as wires, wireless communication links, or fiber optic cables, among others.

The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages and the like. The terminal device 101 may be hardware or software. When the terminal device 101 is hardware, it may be various electronic devices. This includes, but is not limited to, smartphones, tablets, laptops, and desktops, among others. When the terminal device 101 is software, it can be installed in the above-mentioned electronic device. It can be implemented as a plurality of software or software modules, or as a single software or software module. No specific limitation is made here.

The server 103 can provide various services. For example, the server 103 may analyze and process data such as subject-relationship tuples and texts acquired from the terminal device 101, and generate processing results (eg subject-relationship-object triplets).

It should be noted that the server 103 may be hardware or software. When the server 103 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server 103 is software, it may be implemented as multiple software or software modules (for example, for providing distributed services), or as a single software or software module. No specific limitation is made here.

It should be noted that the method for generating information provided in the embodiment of the present application is generally executed by the server 103 , and correspondingly, the device for generating information is generally disposed in the server 103 .

It should be understood that the numbers of terminal devices, networks and servers in Fig. 1 are only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers.

Continue referring to FIG. 2 , which shows a flow 200 of an embodiment of a method for generating information according to the present application. The method for generating information includes the following steps:

Step 201, acquire subject-relationship tuples and text.

In this embodiment, the subject of execution of the method for generating information (for example, the server 103 shown in FIG. 1 ) can acquire subject-relationship tuples and text. Wherein, the subject-relationship dyad may include a subject and a relationship, and is also called an SP dyad. The text can be the information describing the subject-relationship dyad, and its content includes not only the subject and the relation, but also the attribute value of the relation, that is, the object.

Step 202, segment the text into text word sequences.

In this embodiment, the above execution subject may segment the text into text word sequences. Usually, the above-mentioned executive body can use the word segmentation technology to segment the text into word granularities to obtain text word sequences.

Step 203, input the subject-relationship pair and the text word sequence into the pre-trained slot-filling model, and obtain the tagging result of the text word sequence.

In this embodiment, the execution subject can input the subject-relationship pair and the text word sequence into the pre-trained slot-filling model to obtain the tagging result of the text word sequence. Among them, the slot-filling model can be used to label the objects in the text word sequence, so the labeling result will at least mark the objects in the text word sequence.

Here, the slot-filling model can use the slot-filling technique to extract the object corresponding to the subject-relationship pair from the text given the text and the subject-relationship pair. The slot-filling model adds the location of subjects and relations, as well as the semantic similarity of relations, and is a sequence labeling model. The slot-filling model inputs subject-relationship dyads and text word sequences, and outputs the labeling results for each word in the text word sequence. According to the labeling results, the words belonging to the object corresponding to the subject-relational dyad in the text word sequence can be identified.

In some optional implementations of this embodiment, the slot-filling model may use a BIOES sequence tagging method to tag text word sequences. Among them, B is begin, indicating the starting position of the object labeling of multiple words. I is inside, which means the middle part of the object labeling of multiple words. O is outside, which means the part that is not the object. E is end, indicating the end position of the object annotation of multiple words. S is single, which means the object of a single word.

Step 204, based on the subject-relationship double-tuple and the labeling result, generate subject-relationship-object triplet.

In this embodiment, the execution subject may generate subject-relationship-object triplets based on the tagging results of subject-relationship doublets and text word sequences. Among them, the subject-relation-object triplet can include the subject, relation and object of the text. Specifically, the above-mentioned execution subject can determine the object according to different tags in the tagging results of the text word sequence, and then combine the objects into the subject-relationship double-tuple to generate the subject-relationship-object triplet. For example, the above-mentioned execution subject may extract words marked as B, I, and E as objects, or extract words marked as S as objects from the labeling results of text word sequences.

Generally, a large number of subject-object-object triples are generated by using the method provided in this embodiment, which can be used to construct a knowledge graph, improve the connectivity of the knowledge graph, and efficiently improve the knowledge richness of the knowledge graph. In addition, from the application point of view, the subject-object triplet can directly meet the user's search needs for knowledge. When a user searches for a subject-relationship dyad, relevant information of the corresponding object can be directly given, effectively improving the efficiency of user retrieval and browsing. When a user searches for a subject, relevant information of the corresponding object can also be given, which greatly enriches the content of the information pushed to the user.

The method and device for generating information provided by the embodiments of the present application firstly obtain subject-relationship dyads and texts; then text is segmented into text word sequences; then subject-relationship dyads and text word sequences are input to the pre-trained The slot-filling model is used to obtain the tagging results of the text word sequence; finally, based on the subject-relationship doublets and the tagging results, the subject-relationship-object triplets are generated. The object is recognized based on the slot-filling model, which retains the sum structure of the subject and the relationship, and improves the accuracy of object recognition.

Further referring to FIG. 3 , it shows a flow 300 of still another embodiment of the method for generating information according to the present application. The slot filling model in the method for generating information may include an input layer, a localization layer, an embedding layer, an encoding layer, a decoding layer and an output layer. Wherein, the encoding layer may include a first Bidirectional Long Short Term Memory network (Bi-LSTM, Bidirectional Long Short Term Memory). The decoding layer may include a positional attention module, a relational attention module and a second bidirectional long short-term memory network. Specifically, the method for generating information includes the following steps:

Step 301, acquire subject-relationship tuples and text.

Step 302, segment the text into text word sequences.

In this embodiment, the specific operations of steps 301-302 have been introduced in detail in steps 201-202 in the embodiment shown in FIG. 2 , and will not be repeated here.

Step 303, input the subject-relationship pair and the text word sequence to the input layer to obtain word sequence features and distance features.

In this embodiment, the executing subject of the method for generating information (for example, the server 103 shown in FIG. 1 ) can input subject-relationship tuples and text word sequences to the input layer to obtain word sequence features and distance features. Among them, the input layer can be used to extract the features of subject-relational dyads and text word sequences. The word sequence feature may include but not limited to at least one of the following: text word sequence, part-of-speech sequence of the text word sequence, named entity recognition sequence of the text word sequence, relational word sequence of the relation, and the like. The distance feature may include but not limited to at least one of the following: the distance from a word in a text word sequence to a subject, the distance from a word in a text word sequence to a relation, and the like.

Step 304, input the distance feature to the positioning layer to obtain position information.

In this embodiment, the above-mentioned execution subject may input the distance feature into the positioning layer to obtain the position information. Among them, the positioning layer is connected to the output layer, and can locate the position information of the words in the text word sequence based on the distance from the word in the text word sequence to the subject, the distance from the word to the relationship in the text word sequence, and other features.

Step 305, input word sequence features and position information to the embedding layer to obtain word sequence vectors and position vectors.

In this embodiment, the above-mentioned execution subject may input word sequence features and position information into the embedding layer to obtain word sequence vectors and positions. Among them, the embedding layer is connected to the output layer and the positioning layer respectively, which can not only vectorize the word sequence features such as the text word sequence, the part-of-speech sequence of the text word sequence, the named entity recognition sequence of the text word sequence, and the relational word sequence of the relationship, but also The position information of the words in the text word sequence can be vectorized.

It should be noted that the vector matrix for vectorizing word sequence features in the embedding layer is different from the vector matrix for vectorizing positional features.

In step 306, the word sequence vector is input to the encoding layer to obtain an encoding vector.

In this embodiment, the execution subject may input the word sequence vector into the encoding layer to obtain the encoding vector. Among them, the encoding layer is connected with the embedding layer, which can encode the word sequence vector. Since the encoding layer includes the first bidirectional long-short-term memory network, the encoding layer can use the first bidirectional long-short-term memory network to encode the word sequence vector.

In step 307, the concatenation of the position vector and the encoding vector is input to the position attention module to obtain the position information of the distance subject and relationship between words in the text word sequence.

In this embodiment, the above-mentioned execution subject can firstly concatenate the position vector and the encoding vector, and then input the concatenated vector to the position attention module to obtain the position information of the distance subject and the relationship between words in the text word sequence.

Here, the execution subject may input the position vector and the encoding vector to the decoding layer to obtain the decoding vector. In order to add more information at the input end, the encoding layer introduces a long-short-term memory network encoder with an attention mechanism. That is, the decoding layer is composed of a position attention module, a relation attention module and a second bidirectional LSTM network. Among them, the position attention module is connected to the embedding layer and the encoding layer respectively, and the position attention of the subject and the relationship can be added in the slot filling model, that is, the position information of the distance between the subject and the relationship between the words in the text word sequence is used as different weights added to the slot-filling model. The relationship attention module is connected to the encoding layer, which can be used to calculate the importance of the relationship between each word in the input text word sequence and the subject relationship binary group, so that more words related to the relationship can be passed in during the decoding process. Information. The second two-way long-short-term memory network is connected to the position attention module, the relationship attention module and the encoding layer respectively, and the encoding vector obtained by the input encoding layer, the word distance subject and the position information of the relationship in the text word sequence obtained by the position attention module And the semantic similarity between the words in the text word sequence and the relationship obtained by the relationship attention module, and output the decoding vector of the words in the text word sequence.

Step 308, input the long-short-term memory network encoding and the encoding vector of the relation into the relational attention module, and obtain the semantic similarity between the word and the relation in the text word sequence.

In this embodiment, the executive body can input the long-short-term memory network encoding and encoding vector of the relationship into the relationship attention module to obtain the semantic similarity between the words in the text word sequence and the relationship. Specifically, the above-mentioned execution subject can first perform long-short-term memory network encoding on the relationship in the subject-relationship tuple to obtain the long-short-term memory network encoding of the relationship, and then encode the long-short-term memory network encoding of the relationship and the obtained encoding vector of the encoding layer Input to the relationship attention module together to get the semantic similarity between the words and the relationship in the text word sequence.

Step 309, input the encoding vector, the position information of the distance between the words in the text word sequence and the relationship between the subject and the relationship, and the semantic similarity between the words and the relationship in the text word sequence to the second bidirectional long-short-term memory network to obtain a decoding vector.

In this embodiment, the above-mentioned execution subject can input the encoding vector, the word distance subject and the position information of the relationship in the text word sequence and the semantic similarity between the word and the relationship in the text word sequence to the second bidirectional long-short-term memory network, Get the decoded vector.

Step 310, input the decoded vector to the output layer to obtain the labeling result.

In this embodiment, the above execution subject may input the decoding vector to the output layer to obtain the labeling result. Wherein, the output layer is connected to the second bidirectional long short-term memory network in the decoding layer.

In some optional implementations of this embodiment, the above-mentioned executive body can perform multi-classification on the decoding vectors of the words in the text word sequence through the activation function, and obtain that the words in the text word sequence belong to each of the multiple categories The probability of the text word sequence is tagged based on the category corresponding to the maximum probability of the word in the text word sequence, and a tagging result is generated. Among them, multi-category is to calculate the probability that a word belongs to each category in multiple categories. The activation function can be Softmax, for example. Softmax can perform multi-classification on the decoding vector of each word in the text word sequence, and represent the relative probability between different categories, that is, score the probabilities of the BIOES labels separately, and then take one of them The label with the highest probability, so as to complete the labeling of each word in the sequence.

Step 311 , generate subject-relation-object triplets based on the subject-relationship pair and the labeling result.

In this embodiment, the specific operation of step 311 has been introduced in detail in step 204 in the embodiment shown in FIG. 2 , and will not be repeated here.

For ease of understanding, Fig. 4 shows a schematic diagram of the structure of the slot-filling model. As shown in Figure 4, the slot-filling model may include an input layer, a localization layer, an embedding layer, an encoding layer, a decoding layer, and an output layer. Wherein, the encoding layer may include a first Bi-LSTM, and the decoding layer may include a position attention module, a relation attention module and a second Bi-LSTM. If you want to extract the release time of the TV series "XX", you can combine the subject "XX" and the relationship "release time" into a relational dyad <XX, release time>, and obtain the description relation dyad <XX, release time >The text of "XX, the play was launched on August 10, 2015". The text "XX, the play was launched on August 10, 2015" can be segmented into the text word sequence "XX, the play was launched on August 10, 2015". The subject relationship dyad <XX, release time> and the text word sequence "XX, the play was launched on August 10, 2015" are input into the slot filling model, and the corresponding labeling result "O, O, O, O, B , I, E, O". It can be seen that the words "2015", "August" and "10th" in the text word sequence are marked as the object of the subject relationship dyad <XX, release time>, so the generated subject relationship object triplet is <XX, release time, August 10, 2015>.

It can be seen from FIG. 3 that, compared with the embodiment corresponding to FIG. 2 , the flow 300 of the method for generating information in this embodiment highlights the structure of the slot filling model. Therefore, the solution described in this embodiment adds the location attention module and the relationship attention module to the slot-filling model, thereby increasing the location of the subject and the relationship, as well as the semantic similarity of the relationship and other information, and further improving the accuracy of object recognition .

Further referring to FIG. 5 , as an implementation of the methods shown in the above figures, the present application provides an embodiment of a device for generating information, the device embodiment corresponds to the method embodiment shown in FIG. 2 , The device can be specifically applied to various electronic devices.

As shown in FIG. 5 , the apparatus 500 for generating information in this embodiment may include: an acquiring unit 501 , a segmentation unit 502 , a labeling unit 503 and a generating unit 504 . Wherein, the acquisition unit 501 is configured to acquire the subject-relationship pair and the text, wherein the subject-relationship pair includes a subject and a relationship; the segmentation unit 502 is configured to segment the text into text word sequences; the labeling unit 503, It is configured to input the subject-relationship pair and the text word sequence into the pre-trained slot filling model to obtain the labeling result of the text word sequence, wherein the slot filling model is used to mark the object in the text word sequence; the generating unit 504 is It is configured to generate a subject-relation-object triplet based on the subject-relationship pair and the labeling result, wherein the subject-relationship-object triplet includes the subject, relation and object of the text.

In this embodiment, in the apparatus 500 for generating information: the specific processing of the acquisition unit 501, the segmentation unit 502, the labeling unit 503, and the generation unit 504 and the technical effects brought about by them can refer to the corresponding embodiment in FIG. 2 respectively. Relevant descriptions of steps 201-204 in , will not be repeated here.

In some optional implementation manners of this embodiment, the slot-filling model includes an input layer, a positioning layer, an embedding layer, an encoding layer, a decoding layer, and an output layer.

In some optional implementations of this embodiment, the labeling unit 503 includes: an input subunit (not shown in the figure), configured to input the subject-relationship pair and the text word sequence into the input layer to obtain the word sequence feature and distance features; the positioning subunit (not shown in the figure), is configured to input the distance feature to the positioning layer to obtain position information; the embedding subunit (not shown in the figure), is configured to use the word sequence feature and The location information is input to the embedding layer to obtain the word sequence vector and the position vector; the coding subunit (not shown in the figure) is configured to input the word sequence vector to the coding layer to obtain the coding vector; the decoding subunit (not shown in the figure) output), configured to input the position vector and encoding vector to the decoding layer to obtain the decoding vector; the output subunit (not shown in the figure) is configured to input the decoding vector to the output layer to obtain the labeling result.

In some optional implementations of this embodiment, the encoding layer includes a first bidirectional long-short-term memory network, and the decoding layer includes a position attention module, a relation attention module, and a second bidirectional long-short-term memory network.

In some optional implementations of this embodiment, the encoding subunit is further configured to: input the concatenation of the position vector and the encoding vector to the position attention module to obtain the position information of the distance subject and relationship between words in the text word sequence ; Input the long-short-term memory network encoding and encoding vector of the relationship into the relationship attention module to obtain the semantic similarity between the words in the text word sequence and the relationship; the encoding vector, the word distance in the text word sequence and the location information of the relationship The semantic similarity between words and relations in the text word sequence is input to the second bidirectional long-short-term memory network to obtain a decoding vector.

In some optional implementations of this embodiment, the output subunit is further configured to: use an activation function to perform multi-classification on the decoding vectors of the words in the text word sequence, and obtain that the words in the text word sequence belong to multiple categories The probability of each category of , wherein, multi-classification is to calculate the probability that words belong to each category in multiple categories; based on the category corresponding to the maximum probability of words in the text word sequence, the text word sequence is marked to generate the labeling result.

In some optional implementations of this embodiment, the word sequence features include at least one of the following: text word sequences, part-of-speech sequences of text word sequences, named entity recognition sequences of text word sequences and relational word sequences of relations, distance features At least one of the following is included: the distance from the word in the text word sequence to the subject, and the distance from the word in the text word sequence to the relation.

In some optional implementations of this embodiment, the slot-filling model uses a BIOES sequence tagging method to tag text word sequences.

Referring now to FIG. 6 , it shows a schematic structural diagram of a computer system 600 suitable for implementing the electronic device (such as the server 103 shown in FIG. 1 ) of the embodiment of the present application. The electronic device shown in FIG. 6 is only an example, and should not limit the functions and scope of use of this embodiment of the present application.

As shown in FIG. 6 , a computer system 600 includes a central processing unit (CPU) 601 that can be programmed according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage section 608 into a random-access memory (RAM) 603 Instead, various appropriate actions and processes are performed. In the RAM 603, various programs and data necessary for the operation of the system 600 are also stored. The CPU 601 , ROM 602 , and RAM 603 are connected to each other via a bus 604 . An input/output (I/O) interface 605 is also connected to the bus 604 .

The following components are connected to the I/O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage section 608 including a hard disk, etc. and a communication section 609 including a network interface card such as a LAN card, a modem, or the like. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I/O interface 605 as needed. A removable medium 611 such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc. is mounted on the drive 610 as necessary so that a computer program read therefrom is installed into the storage section 608 as necessary.

In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, where the computer program includes program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via communication portion 609 and/or installed from removable media 611 . When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the method of the present application are performed.

It should be noted that the computer-readable medium described in this application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, in which computer-readable program codes are carried. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. . Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out the operations of this application can be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, and conventional A procedural programming language—such as "C" or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or electronic device. In cases involving a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as through an Internet service provider). Internet connection).

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or portion of code that contains one or more logical functions for implementing specified executable instructions. 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 they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified functions or operations , or may be implemented by a combination of dedicated hardware and computer instructions.

The units involved in the embodiments described in the present application may be implemented by means of software or by means of hardware. The described units may also be set in a processor, for example, it may be described as: a processor includes an acquisition unit, a segmentation unit, a labeling unit and a generation unit. Wherein, the names of these units do not limit the unit itself in this case, for example, the acquisition unit may also be described as "a unit for acquiring subject-relationship tuples and text".

As another aspect, the present application also provides a computer-readable medium. The computer-readable medium may be included in the electronic device described in the above embodiments; it may also exist independently without being assembled into the electronic device. middle. The above-mentioned computer-readable medium carries one or more programs, and when the above-mentioned one or more programs are executed by the electronic device, the electronic device: acquires the subject-relationship pair and the text, wherein the subject-relationship pair includes subject and relations; split the text into text word sequences; input the subject-relationship dyads and text word sequences into the pre-trained slot filling model to obtain the tagging results of the text word sequences, where the slot filling model is used to mark the text word sequences based on the subject-relationship pair and the labeling results, a subject-relation-object triplet is generated, where the subject-relationship-object triplet includes the subject, relation and object of the text.

The above description is only a preferred embodiment of the present application and an illustration of the applied technical principle. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solution formed by the specific combination of the above-mentioned technical features, and should also cover the technical solution formed by the above-mentioned technical features or Other technical solutions formed by any combination of equivalent features. For example, a technical solution formed by replacing the above-mentioned features with technical features with similar functions disclosed in (but not limited to) this application.

Claims (14)

1. A method for generating information comprising: Obtaining a subject-relationship pair and text, wherein the subject-relationship pair includes subject and relation; Segmenting the text into sequences of text words; Input the subject-relationship pair and the text word sequence into a pre-trained slot-filling model to obtain the tagging result of the text word sequence, wherein the slot-filling model is used to tag the text word sequence object; Based on the subject-relationship pair and the annotation result, a subject-relation-object triplet is generated, wherein the subject-relationship-object triplet includes the subject, relation, and object of the text; Wherein, the slot filling model includes an input layer, a positioning layer, an embedding layer, an encoding layer, a decoding layer and an output layer; and Said inputting said subject-relationship pair and said text word sequence into the pre-trained slot filling model to obtain the labeling result of said text word sequence, including: Input the subject-relationship binary group and the text word sequence to the input layer to obtain word sequence features and distance features; Inputting the distance feature into the positioning layer to obtain position information; The word sequence feature and the position information are input to the embedding layer to obtain a word sequence vector and a position vector; The word sequence vector is input to the encoding layer to obtain an encoding vector; inputting the position vector and the encoding vector to the decoding layer to obtain a decoding vector; The decoding vector is input to the output layer to obtain the labeling result. 2. The method of claim 1, wherein the encoding layer comprises a first bidirectional LSTM network and the decoding layer comprises a positional attention module, a relational attention module and a second bidirectional LSTM network. 3. The method according to claim 2, wherein said inputting said position vector and said encoding vector to said decoding layer to obtain a decoding vector comprises: The concatenation of the position vector and the encoding vector is input to the position attention module to obtain the position information of the words in the text word sequence from the subject and the relationship; The long-short-term memory network encoding of the relationship and the encoding vector are input to the relationship attention module to obtain the semantic similarity between the words in the text word sequence and the relationship; Input the encoding vector, the position information of the words in the text word sequence from the subject and the relationship, and the semantic similarity between the words in the text word sequence and the relationship to the second bidirectional long short-term memory network to obtain the decoded vector. 4. The method according to claim 1, wherein said inputting said decoding vector to said output layer to obtain said labeling result comprises: The decoding vectors of the words in the text word sequence are multi-classified by the activation function to obtain the probability that the words in the text word sequence belong to each category in multiple categories, wherein the multi-classification is to calculate the word belonging to multiple categories the probability of each of the categories; The text word sequence is tagged based on the category corresponding to the maximum probability of the word in the text word sequence, and the tagging result is generated. 5. The method according to one of claims 1-4, wherein the word sequence feature comprises at least one of the following: the text word sequence, the part-of-speech sequence of the text word sequence, the naming of the text word sequence An entity recognition sequence and a relational word sequence of the relationship, the distance feature includes at least one of the following: the distance from the word in the text word sequence to the subject, the distance from the word in the text word sequence to the relationship distance. 6. The method according to any one of claims 1-4, wherein the slot-filling model uses a BIOES sequence tagging method to tag the text word sequence. 7. An apparatus for generating information comprising: An acquisition unit configured to acquire a subject-relationship tuple and text, wherein the subject-relationship tuple includes a subject and a relationship; a segmentation unit configured to segment the text into text word sequences; A labeling unit configured to input the subject-relationship pair and the text word sequence into a pre-trained slot-filling model to obtain a labeling result of the text word sequence, wherein the slot-filling model is used to label the object in the sequence of words in the prescriptive text; A generating unit configured to generate a subject-relation-object triplet based on the subject-relationship pair and the annotation result, wherein the subject-relationship-object triplet includes the subject, relation, and object of the text; Wherein, the slot filling model includes an input layer, a positioning layer, an embedding layer, an encoding layer, a decoding layer and an output layer; and The labeling unit includes: The input subunit is configured to input the subject-relationship pair and the text word sequence to the input layer to obtain word sequence features and distance features; a positioning subunit configured to input the distance feature to the positioning layer to obtain position information; An embedding subunit configured to input the word sequence feature and the position information to the embedding layer to obtain a word sequence vector and a position vector; An encoding subunit configured to input the word sequence vector to the encoding layer to obtain an encoding vector; a decoding subunit configured to input the position vector and the encoding vector to the decoding layer to obtain a decoding vector; The output subunit is configured to input the decoding vector to the output layer to obtain the labeling result. 8. The apparatus of claim 7, wherein the encoding layer comprises a first bidirectional LSTM network and the decoding layer comprises a positional attention module, a relational attention module and a second bidirectional LSTM network. 9. The apparatus according to claim 8, wherein the encoding subunit is further configured to: The concatenation of the position vector and the encoding vector is input to the position attention module to obtain the position information of the words in the text word sequence from the subject and the relationship; The long-short-term memory network encoding of the relationship and the encoding vector are input to the relationship attention module to obtain the semantic similarity between the words in the text word sequence and the relationship; Input the encoding vector, the position information of the words in the text word sequence from the subject and the relationship, and the semantic similarity between the words in the text word sequence and the relationship to the second bidirectional long short-term memory network to obtain the decoded vector. 10. The apparatus of claim 7, wherein the output subunit is further configured to: The decoding vectors of the words in the text word sequence are multi-classified by the activation function to obtain the probability that the words in the text word sequence belong to each category in multiple categories, wherein the multi-classification is to calculate the word belonging to multiple categories the probability of each of the categories; The text word sequence is tagged based on the category corresponding to the maximum probability of the word in the text word sequence, and the tagging result is generated. 11. The device according to any one of claims 7-10, wherein the word sequence features include at least one of the following: the text word sequence, the part-of-speech sequence of the text word sequence, the naming of the text word sequence An entity recognition sequence and a relational word sequence of the relationship, the distance feature includes at least one of the following: the distance from the word in the text word sequence to the subject, the distance from the word in the text word sequence to the relationship distance. 12. The device according to any one of claims 7-10, wherein the slot-filling model uses a BIOES sequence tagging method to tag the text word sequence. 13. An electronic device comprising: one or more processors; a storage device on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method according to any one of claims 1-6. 14. A computer-readable medium, on which a computer program is stored, wherein the computer program implements the method according to any one of claims 1-6 when executed by a processor.

Images