CN111368066B - Method, apparatus and computer readable storage medium for obtaining dialogue abstract - Google Patents
Method, apparatus and computer readable storage medium for obtaining dialogue abstract Download PDFInfo
- Publication number
- CN111368066B CN111368066B CN201811483302.3A CN201811483302A CN111368066B CN 111368066 B CN111368066 B CN 111368066B CN 201811483302 A CN201811483302 A CN 201811483302A CN 111368066 B CN111368066 B CN 111368066B
- Authority
- CN
- China
- Prior art keywords
- dialogue
- abstract
- sentences
- information
- sentence
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000010801 machine learning Methods 0.000 claims abstract description 34
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 50
- 239000013598 vector Substances 0.000 claims description 31
- 238000004590 computer program Methods 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 4
- 238000005516 engineering process Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 16
- 230000006870 function Effects 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000002372 labelling Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 230000011218 segmentation Effects 0.000 description 4
- 230000000306 recurrent effect Effects 0.000 description 3
- 230000002457 bidirectional effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- JEIPFZHSYJVQDO-UHFFFAOYSA-N iron(III) oxide Inorganic materials O=[Fe]O[Fe]=O JEIPFZHSYJVQDO-UHFFFAOYSA-N 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000009466 transformation 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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
The present disclosure relates to a method, an apparatus, and a computer-readable storage medium for obtaining a conversation digest, and relates to the field of computer technology. The method of the present disclosure comprises: acquiring dialogue information; determining respective dialogue abstracts of each dialogue participant from dialogue information by utilizing a pre-trained machine learning model; and matching related abstract contents from the dialogue abstracts of all the dialogue participants according to semantic similarity, and generating dialogue abstract combinations as abstracts of dialogue information. The method disclosed by the invention is applied to the abstract extraction process of the dialogue between the client and the customer service of the electronic commerce platform, and can automatically extract the core problems of the client and the replies of the customer service respectively, so that the problems of the client and the replies of the customer service are matched, an accurate and clear dialogue abstract is formed, and the efficiency of conducting the dialogue abstract is improved.
Description
Technical Field
The present disclosure relates to the field of computer technology, and in particular, to a method, an apparatus, and a computer readable storage medium for obtaining a session abstract.
Background
With the rise of the electronic commerce platform, the service efficiency, experience and quality requirements of consumers on online customer service of the electronic commerce platform are continuously improved, and higher requirements are provided for a customer service system of the electronic commerce platform. For highly specialized or complex business, a customer usually needs to communicate with a plurality of customer service personnel or machines in order to solve the problem. Therefore, each customer service needs to summarize dialogue contents to generate a text abstract after knowing the requirements of the customer, so that the customer service can quickly know the requirements of the customer conveniently, and the repeated explanation of the requirements of the customer and the customer service can be avoided. For example, solutions are given or forwarded to other customer services based on the logistics, product quality, shipping errors, etc. presented by the customer. When the call needs to be transferred to other customer service, the other customer service needs to know the call content before the customer as a reference. In this process, the accuracy, generalization and instantaneity of the text abstract are particularly important.
At present, the conversation abstract of customer service and customer is completed by manual customer service, and after each online conversation or voice conversation is finished, the manual customer service also needs 1-2 minutes to complete the summary of the conversation service content, and the core problem of customer and the reply of customer service are extracted.
Disclosure of Invention
The inventors found that: because of the specificity of the conversation form of the conversation, the extraction of the abstract of the conversation between the customer and the customer service needs to be completed manually, namely, the core problem of the customer and the reply of the customer service are extracted. However, this method of summarization is inefficient, and a method for automatically extracting the dialog summary is needed to improve the efficiency of performing the dialog summary.
One technical problem to be solved by the present disclosure is: a method for automatically extracting dialogue abstracts is provided, so that the efficiency of dialogue abstracts is improved.
According to some embodiments of the present disclosure, a method for obtaining a conversation abstract is provided, including: acquiring dialogue information; determining respective dialogue abstracts of each dialogue participant from dialogue information by utilizing a pre-trained machine learning model; and matching related abstract contents from the dialogue abstracts of all the dialogue participants according to semantic similarity, and generating dialogue abstract combinations as abstracts of dialogue information.
In some embodiments, determining respective conversation summaries for respective conversation participants from conversation information using a pre-trained machine learning model includes: dividing dialogue information into dialogue information of each dialogue participant according to the identification of the dialogue participant; respectively inputting each sentence in the dialogue information of each dialogue participant into a pre-trained machine learning model to determine the probability that each sentence corresponding to the dialogue participant belongs to the abstract sentence; and selecting sentences as dialogue summaries of the dialogue participants according to the probabilities that the sentences corresponding to the dialogue participants belong to the summary sentences.
In some embodiments, matching relevant digest content from the conversation digests of the individual conversation participants based on semantic similarity, generating a conversation digest combination includes: combining sentences in the dialogue abstracts of different dialogue participants two by two to obtain all candidate abstract combinations; determining whether sentences in each candidate abstract combination are semantically similar; and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
In some embodiments, determining whether sentences in each candidate summary combination are semantically similar comprises: mapping sentences in the candidate abstract combination into vectors according to a preset rule; and determining whether the two sentences are similar according to the distance between the vectors of the two sentences in the candidate abstract combination.
In some embodiments, the method further comprises: acquiring dialogue information marked with dialogue abstract as training sample; the machine learning model is trained using training samples.
In some embodiments, obtaining dialogue information labeled with the dialogue digest as a training sample includes: performing similarity matching on each sentence in the dialogue information for training and the artificially generated dialogue abstract to obtain a matching score; the sentence with the highest matching score is marked as a positive dialogue abstract sample, and other sentences are marked as negative dialogue abstract samples.
According to other embodiments of the present disclosure, there is provided an apparatus for obtaining a conversation digest, including: the dialogue acquisition module is used for acquiring dialogue information; the abstract extraction module is used for respectively determining the dialogue abstracts of each dialogue participant from dialogue information by utilizing a pre-trained machine learning model; and the abstract determining module is used for matching related abstract contents from the dialogue abstracts of all dialogue participants according to semantic similarity and generating dialogue abstract combinations as abstracts of dialogue information.
In some embodiments, the abstract extraction module is configured to divide the dialogue information into dialogue information of each dialogue participant according to the identification of the dialogue participant; respectively inputting each sentence in the dialogue information of each dialogue participant into a pre-trained machine learning model to determine the probability that each sentence corresponding to the dialogue participant belongs to the abstract sentence; and selecting sentences as dialogue summaries of the dialogue participants according to the probabilities that the sentences corresponding to the dialogue participants belong to the summary sentences.
In some embodiments, the summary determining module is configured to combine sentences in the conversation summaries of different conversation participants two by two to obtain all candidate summary combinations; determining whether sentences in each candidate abstract combination are semantically similar; and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
In some embodiments, the summary determination module is configured to map sentences in the candidate summary combination into vectors according to a preset rule; and determining whether the two sentences are similar according to the distance between the vectors of the two sentences in the candidate abstract combination.
In some embodiments, the apparatus further comprises: the training sample generation module is used for acquiring dialogue information marked with the dialogue abstract as a training sample; and the training module is used for training the machine learning model by utilizing the training sample.
In some embodiments, the training sample generation module is configured to perform similarity matching on each sentence in the training dialogue information and the artificially generated dialogue abstract to obtain a matching score; the sentence with the highest matching score is marked as a positive dialogue abstract sample, and other sentences are marked as negative dialogue abstract samples.
According to still further embodiments of the present disclosure, there is provided an apparatus for obtaining a conversation digest, including: a memory; and a processor coupled to the memory, the processor configured to perform the method of retrieving a dialog digest as in any of the embodiments described above based on instructions stored in the memory.
According to still further embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, wherein the program, when executed by a processor, implements the method of obtaining a conversation digest of any of the previous embodiments.
The conversation abstract of each conversation participant is extracted from the conversation information by a pre-trained machine learning model. Further, the respective conversation summaries of the conversation participants are matched and combined according to the semantic similarity, and the conversation summary combination is generated as the summary of the whole conversation. The method disclosed by the invention is applied to the abstract extraction process of the dialogue between the client and the customer service of the electronic commerce platform, and can automatically extract the core problems of the client and the replies of the customer service respectively, so that the problems of the client and the replies of the customer service are matched, an accurate and clear dialogue abstract is formed, and the efficiency of conducting the dialogue abstract is improved.
Other features of the present disclosure and its advantages will become apparent from the following detailed description of exemplary embodiments of the disclosure, which proceeds with reference to the accompanying drawings.
Drawings
In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and other drawings may be obtained according to these drawings without inventive effort to a person of ordinary skill in the art.
Fig. 1 illustrates a flow diagram of a method of retrieving a conversation digest in some embodiments of the present disclosure.
Fig. 2 shows a flow diagram of a method of retrieving a conversation digest in further embodiments of the present disclosure.
Fig. 3 illustrates a schematic diagram of an apparatus for retrieving a conversation digest in accordance with some embodiments of the present disclosure.
Fig. 4 is a schematic diagram illustrating a structure of an apparatus for retrieving a conversation digest according to other embodiments of the present disclosure.
Fig. 5 shows a schematic structural diagram of an apparatus for obtaining a conversation digest according to still other embodiments of the present disclosure.
Fig. 6 shows a schematic structural diagram of an apparatus for retrieving a conversation digest in accordance with still other embodiments of the present disclosure.
Detailed Description
The following description of the technical solutions in the embodiments of the present disclosure will be made clearly and completely with reference to the accompanying drawings in the embodiments of the present disclosure, and it is apparent that the described embodiments are only some embodiments of the present disclosure, not all embodiments. The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses. Based on the embodiments in this disclosure, all other embodiments that a person of ordinary skill in the art would obtain without making any inventive effort are within the scope of protection of this disclosure.
The present disclosure presents an automatic method for obtaining a dialog digest, which is described below in connection with fig. 1.
Fig. 1 is a flow chart of some embodiments of a method of retrieving a conversation digest of the present disclosure. As shown in fig. 1, the method of this embodiment includes: steps S102 to S106.
In step S102, dialogue information is acquired.
The dialogue is applied to a dialogue scene of a customer and customer service of an electronic commerce platform, and the dialogue comprises two participants. The first party and the second party are respectively clients and client service personnel or clients and client service robots in the electronic commerce platform. The schemes of the present disclosure are also applicable to other dialog scenarios, and are not limited to the illustrated examples. The dialogue information includes, for example, text or voice of a dialogue, and the like.
In step S104, a respective conversation digest of each conversation participant is determined from the conversation information using a pre-trained machine learning model.
In the case where the dialogue information is speech, the existing model may be used to convert the speech into text and then determine the dialogue digest. The pre-trained machine learning model is, for example, an existing text summarization model. Further, in the scenario of a customer-customer service dialogue, since the sentence length in the dialogue is generally short, the pre-trained machine learning model can be an existing extraction type text abstract model, i.e. find out sentences which can be used as abstracts from dialogue information.
In some embodiments, the dialogue information is divided into dialogue information of each dialogue participant according to the identification of the dialogue participant; respectively inputting each sentence in the dialogue information of each dialogue participant into a pre-trained machine learning model to determine the probability that each sentence corresponding to the dialogue participant belongs to the abstract sentence; and selecting sentences as dialogue summaries of the dialogue participants according to the probabilities that the sentences corresponding to the dialogue participants belong to the summary sentences.
For example, in the case where two conversation participants are provided, the conversation information is divided into conversation information of a first party and conversation information of a second party according to the identification of the conversation parties; inputting each sentence in the dialogue information of the first party into a pre-trained machine learning model to determine the probability that each sentence corresponding to the first party belongs to the abstract sentence, and inputting each sentence in the dialogue information of the second party into the pre-trained machine learning model to determine the probability that each sentence corresponding to the second party belongs to the abstract sentence; and selecting sentences as the dialogue abstract of the first party and the dialogue abstract of the second party according to the probability that each sentence corresponding to the first party belongs to the abstract sentence and the probability that each sentence corresponding to the second party belongs to the abstract sentence.
The dialogue information is divided into a plurality of parts, each part being dialogue information belonging to one party. For example, in the case where there are two conversation participants, one part is all the conversation information of the first party, and the other part is all the conversation information of the second party. The dialogue information of each dialogue participant is respectively input into a pre-trained machine learning model, so that the dialogue abstract of each dialogue participant output by the pre-trained machine learning model is respectively obtained. Further, for example, each sentence in the dialogue information of each dialogue participant is segmented, a word vector of each segmented word is determined, and the word vector of each segmented word is input into a pre-trained machine learning model.
The embodiment of step S104 will be described below taking a pre-trained machine learning model as SummaRuNNer (A Recurrent Neural Network based Sequence Model for Extractive Summarization of Documents, a document summary extraction model based on a recurrent neural network), and two dialogue participants as examples.
The dialogue information is divided into dialogue information of a first party and dialogue information of a second party. Each sentence in the dialogue information of the first party and the dialogue information of the second party is segmented. A word vector of the word in the dialogue information of the first party and a word vector of the word in the dialogue information of the second party are determined. Word2vec algorithm is used to determine word vectors for word segmentation, for example. And inputting word vectors of the word segmentation in the dialogue information of the first party into the SummaRunner to obtain the probability that each sentence in the dialogue information of the first party belongs to the abstract sentence. Sentences with probabilities greater than the threshold are selected as the dialogue digests of the first party. And inputting word vectors of the word segmentation in the dialogue information of the second party into the SummaRunner to obtain the probability that each sentence in the dialogue information of the second party belongs to the abstract sentence. Sentences with probabilities greater than the threshold are selected as the dialogue digests of the second party. Or the SummaRuNNer may be configured to select a predetermined proportion of sentences from the sentences of the dialogue information as abstract sentences, for example, 1 from the 5 dialogue sentences as abstract sentences.
SummaRunner consists of two layers of bidirectional GRU-RNNs (Gated Recurrent Units-Recurrent Neural Network, gated cyclic units-cyclic neural networks). The first layer of GRU-RNN operates on a word level to calculate the implicit state expression of each word based on its word vector and the previous implicit state. Meanwhile, another RNN is used for word-level operation, the end of the sentence is started to the head of the sentence, and the RNN and the previous RNN are spliced into a bidirectional RNN. The second layer of the model is also composed of a bi-directional RNN, which runs the computation at the sentence level, whose hidden states represent sentences in the text, which takes as input the output of the first layer bi-directional RNN. Eventually, the entire text is represented as a non-linear transformation of the bi-directional RNN at one sentence level. The top layer is a classification layer based on a sigmoid activation function, and is used for determining whether each sentence belongs to a summary sentence or not. Classification depends on the richness of the content of the sentence, its relevance to the document, and the novelty of the sentence with respect to the document. Since the SummaRuNNer belongs to the existing model, the description thereof is omitted.
In step S106, the relevant digest contents are matched from the session digests of the respective session participants according to the semantic similarity, and a session digest combination is generated as a digest of the session information.
For example, in the case where there are two conversation participants, a conversation digest combination is generated as a digest of conversation information from a conversation digest of a first party and a conversation digest of a second party, which match the relevant digest contents.
Because the context sentences in the dialogue have strong general semantic relevance, the dialogue summaries of the two parties of the dialogue can be matched by utilizing the semantic similarity, so that the final abstract of the dialogue information is determined.
In some embodiments, sentences in the dialogue abstracts of different dialogue participants are combined two by two to obtain all candidate abstract combinations; determining whether sentences in each candidate abstract combination are semantically similar; and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
For example, in the case that the number of dialogue participants is two, combining sentences in the dialogue abstract of the first party with sentences in the dialogue abstract of the second party two by two to obtain all candidate abstract combinations; determining whether sentences in each candidate abstract combination are semantically similar; and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
For example, a first party's conversation digest contains two sentences { Q1, Q2}, and a second party's conversation digest contains three sentences { A1, A2, A3}. All candidate summary combinations are obtained as follows,
{ Q1, A1}, { Q1, A2}, { Q1, A3}, { Q2, A1}, { Q2, A2}, { Q2, A3}. And calculating semantic similarity for two sentences in each candidate abstract combination, if the semantic similarity is greater than a threshold value, considering the semantic similarity of the two sentences, and selecting the candidate abstract combination with the similar sentence semantic as a dialogue abstract combination.
In some embodiments, sentences in the candidate summary combinations are mapped into vectors according to preset rules; and determining whether the two sentences are similar according to the distance between the vectors of the two sentences in the candidate abstract combination. I.e. the distance of the vectors of the two sentences is calculated as the semantic similarity of the two sentences.
Existing semantic similarity calculation models can be used to match related summary content from the conversation summaries of the various conversation participants. For example, it may be employed. DSSM (Deep Structured Semantic Models, deep-structure semantic model) matches relevant summary content from the conversation summaries of the individual conversation participants. The principle of the DSSM is that two sentences are respectively expressed into low-dimensional semantic vectors through DNN (Deep Neural Network ), the distance between the two semantic vectors is calculated through a cosine distance, and finally the semantic similarity of the two sentences is determined. The model can be used for predicting the semantic similarity of two sentences and obtaining the low latitude semantic vector expression of a certain sentence. DSSM can be divided into three layers from bottom to top: input layer, presentation layer, matching layer. The input layer does this by mapping sentences into a vector space and inputting into the DNN. The DSSM representation layer adopts a BOW (Bag of words) mode, which is equivalent to discarding the position information of the word vector, and the words in the whole sentence are placed in one Bag and are not sequenced. Finally, the semantic similarity of two sentences can be represented by the cosine distance of the two semantic vectors. The semantic similarity between two sentences can be converted into a posterior probability through the softmax function, and whether the two sentences are semantically similar or not is further determined according to the probability. DSSM belongs to the existing model and will not be described in detail here.
It should be noted that, the model for determining the dialogue digest of each dialogue participant and the model for matching the related digest content from the dialogue digest of each dialogue participant may be any existing text digest model and text similarity calculation model, and is not limited to the above-mentioned examples.
In the method of the above embodiment, the dialogue abstracts of the dialogue participants are extracted from the dialogue information respectively by using a pre-trained machine learning model. Further, the dialogue summaries of all dialogue participants are matched and combined according to the semantic similarity, and dialogue summary combination is generated as the summary of the whole dialogue. The method of the embodiment is applied to the abstract extraction process of the dialogue between the client and the customer service of the electronic commerce platform, and can automatically extract the core problems of the client and the replies of the customer service respectively, so that the problems of the client and the replies of the customer service are matched, an accurate and clear dialogue abstract is formed, and the efficiency of conducting the dialogue abstract is improved.
The present disclosure also provides some embodiments of training a model, as described below in connection with fig. 2.
Fig. 2 is a flow chart of some embodiments of a method of retrieving a conversation digest of the present disclosure. As shown in fig. 2, the method of this embodiment includes: steps S202 to S204.
In step S202, dialogue information labeled with a dialogue digest is acquired as a training sample.
For the scene of the dialogue between the customer and the customer service, the content of each dialogue can be stored into a database in the electronic commerce platform, and the unique number is marked, so that the content of the dialogue can be found through the number. The staff member in the prior art can summarize the dialog digest for each dialog and store it as a database. Training samples may be generated using data in a database. However, since the manually summarized dialogue abstract may not be the original sentence in the dialogue, a training sample is required to be obtained after a certain process is performed in combination with the original dialogue content and the manually summarized dialogue abstract.
In some embodiments, similarity matching is performed on each sentence in the training dialogue information with the artificially generated dialogue abstract to obtain a matching score; the sentence with the highest matching score is marked as a positive dialogue abstract sample, and other sentences are marked as negative dialogue abstract samples. For example, a segment of dialogue information contains 5 sentences, the artificially generated dialogue abstract is 1 sentence, then the matching score is calculated between the 5 sentences in the dialogue information and the 1 sentence of the artificially generated dialogue abstract, the sentence with the highest matching score is selected to be marked with a label (for example, 1) to represent the sentence as a positive sample, and other sentences are marked with a label (for example, 0) to represent the sentence as a negative sample.
Further, each sentence in the dialogue information and the artificially generated dialogue abstract may be divided into a plurality of words according to a preset number of characters, and the matching score may be determined according to the number of the same words appearing in each sentence in the dialogue information and the artificially generated dialogue abstract. For example, a Rouge value of each sentence in the training dialogue information and the manually generated dialogue digest may be calculated as the matching score. Other automatic digest evaluation methods may also be used to calculate a match score for each sentence in the dialogue information to the artificially generated dialogue digest, not limited to the illustrated example.
In some embodiments, the dialogue information is divided into respective dialogue information of dialogue participants according to the identification of the dialogue participants; the dialogue information of each participant can be used as a training sample.
In step S204, the machine learning model is trained using the training samples.
Inputting each sentence of the training sample into a machine learning model to output the probability that each sentence belongs to the abstract sentence; calculating a loss function according to the output probability and the labeling information; and adjusting parameters of the machine learning model until the loss function value reaches the minimum, and finishing training.
For example, if the machine learning model is SummaRuNNer, word segmentation is performed on each sentence of the training sample, and word vectors of each segmented word are determined; inputting word vectors of training samples into SummaRunner, and outputting probabilities that each sentence belongs to a summary sentence; and calculating Cross Entropy (Cross Entropy) according to the output probability and the labeling information, and adjusting parameters of SummaRunner until the Cross Entropy is minimum, thereby completing training.
According to the method, the training samples can be automatically generated by utilizing the existing data, the process of manual repeatability labeling is reduced, and the training sample generation efficiency and the training efficiency of the model are improved.
The semantic similarity determination model may also be trained using manually annotated dialog digest combinations. For example, the dialogue abstracts have been extracted from the training dialogue information, and the sentence combinations of the semantic similarity in the dialogue abstracts of the two parties can be manually labeled as positive samples, and the sentence combinations of the dialogue abstracts of the other two parties can be manually labeled as negative samples. Inputting each sentence combination into a semantic similarity determination model to output the matching score of the sentences in the combination; calculating a loss function according to the output matching score and the labeling information; and adjusting parameters of the semantic similarity determination model until the loss function value reaches the minimum, and completing training.
For example, the semantic similarity determining model is DSSM, each labeled sentence is input into DSSM in a combined mode, so that vectors of two sentences are obtained, the cosine distance of the two vectors is calculated, a posterior probability of semantic similarity conversion between the two sentences is output, and a loss function is calculated according to the output probability and labeling information; and adjusting parameters of the DSSM until the loss function value reaches the minimum, and finishing training.
The present disclosure also provides an apparatus for obtaining a conversation digest, which is described below in conjunction with fig. 3.
Fig. 3 is a block diagram of some embodiments of an apparatus of the present disclosure for retrieving a conversation digest. As shown in fig. 3, the apparatus 30 of this embodiment includes: the dialogue acquisition module 302, the digest extraction module 304, and the digest determination module 306.
The dialogue acquisition module 302 is configured to acquire dialogue information.
The abstract extraction module 304 is configured to determine respective session summaries of the session participants from the session information using a pre-trained machine learning model.
In some embodiments, the abstract extraction module 304 is configured to divide the dialogue information into dialogue information of each dialogue participant according to the identification of the dialogue participant; respectively inputting each sentence in the dialogue information of each dialogue participant into a pre-trained machine learning model to determine the probability that each sentence corresponding to the dialogue participant belongs to the abstract sentence; and selecting sentences as dialogue summaries of the dialogue participants according to the probabilities that the sentences corresponding to the dialogue participants belong to the summary sentences.
The digest determining module 306 is configured to match related digest contents from the session digests of the session participants according to the semantic similarity, and generate a session digest combination as a digest of the session information.
In some embodiments, the summary determining module 306 is configured to combine sentences in the conversation summaries of different conversation participants two by two to obtain all candidate summary combinations; determining whether sentences in each candidate abstract combination are semantically similar; and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
In some embodiments, the summary determination module 306 is configured to map sentences in the candidate summary combination into vectors according to a preset rule; and determining whether the two sentences are similar according to the distance between the vectors of the two sentences in the candidate abstract combination.
Further embodiments of the apparatus for retrieving a conversation digest of the present disclosure are described below in conjunction with fig. 4.
Fig. 4 is a block diagram of further embodiments of an apparatus for retrieving a conversation digest according to the present disclosure. As shown in fig. 4, the apparatus 40 of this embodiment includes: the functions of the dialogue acquisition module 402, the abstract extraction module 404, and the abstract determination module 406 are the same as or similar to those of the dialogue acquisition module 302, the abstract extraction module 304, and the abstract determination module 306, respectively; and a training sample generation module 408, a training module 410.
The training sample generation module 408 is configured to obtain dialogue information labeled with the dialogue abstract as a training sample.
In some embodiments, the training sample generation module 408 is configured to perform similarity matching on each sentence in the training dialogue information and the artificially generated dialogue abstract to obtain a matching score; the sentence with the highest matching score is marked as a positive dialogue abstract sample, and other sentences are marked as negative dialogue abstract samples.
The training module 410 is configured to train the machine learning model using the training samples.
The apparatus for retrieving a conversation digest in embodiments of the present disclosure may each be implemented by various computing devices or computer systems, as described below in connection with fig. 5 and 6.
Fig. 5 is a block diagram of some embodiments of an apparatus of the present disclosure for retrieving a conversation digest. As shown in fig. 5, the apparatus 50 of this embodiment includes: a memory 510 and a processor 520 coupled to the memory 510, the processor 520 being configured to perform the method of retrieving a conversation digest in any of the embodiments of the present disclosure based on instructions stored in the memory 110.
The memory 510 may include, for example, system memory, fixed nonvolatile storage media, and the like. The system memory stores, for example, an operating system, application programs, boot Loader (Boot Loader), database, and other programs.
Fig. 6 is a block diagram of further embodiments of an apparatus for retrieving a conversation digest according to the present disclosure. As shown in fig. 6, the apparatus 60 of this embodiment includes: memory 610 and processor 620 are similar to memory 510 and processor 520, respectively. Input/output interface 630, network interface 640, storage interface 650, and the like may also be included. These interfaces 630, 640, 650 and the memory 610 and processor 620 may be connected by, for example, a bus 660. The input/output interface 630 provides a connection interface for input/output devices such as a display, a mouse, a keyboard, and a touch screen. The network interface 640 provides a connection interface for various networking devices, such as may be connected to a database server or cloud storage server, or the like. The storage interface 650 provides a connection interface for external storage devices such as SD cards, U-discs, and the like.
It will be appreciated by those skilled in the art that embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each flowchart and/or block of the flowchart illustrations and/or block diagrams, and combinations of flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing description of the preferred embodiments of the present disclosure is not intended to limit the disclosure, but rather to enable any modification, equivalent replacement, improvement or the like, which fall within the spirit and principles of the present disclosure.
Claims (12)
1. A method of obtaining a conversation digest, comprising:
acquiring dialogue information;
determining respective conversation summaries of each conversation participant from the conversation information by utilizing a pre-trained machine learning model;
matching related abstract content from the dialogue abstracts of all dialogue participants according to semantic similarity, and generating dialogue abstract combinations as abstracts of the dialogue information;
wherein the matching related abstract content from the dialogue abstracts of the dialogue participants according to semantic similarity, and generating dialogue abstract combination comprises:
combining sentences in the dialogue abstracts of different dialogue participants two by two to obtain all candidate abstract combinations;
determining whether sentences in each candidate abstract combination are semantically similar;
and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
2. The method for retrieving a conversation digest of claim 1 wherein,
the determining, from the dialogue information, respective dialogue summaries of respective dialogue participants using a pre-trained machine learning model includes:
dividing the dialogue information into dialogue information of each dialogue participant according to the identification of the dialogue participant;
respectively inputting each sentence in the dialogue information of each dialogue participant into a pre-trained machine learning model to determine the probability that each sentence corresponding to the dialogue participant belongs to the abstract sentence;
and selecting sentences as dialogue summaries of the dialogue participants according to the probabilities that the sentences corresponding to the dialogue participants belong to the summary sentences.
3. The method for retrieving a conversation digest of claim 1 wherein,
the determining whether sentences in each candidate abstract combination are semantically similar comprises the following steps:
mapping sentences in the candidate abstract combination into vectors according to a preset rule;
and determining whether the two sentences are similar according to the distance between the vectors of the two sentences in the candidate abstract combination.
4. A method of retrieving a conversation digest as claimed in any one of claims 1 to 3 further comprising:
acquiring dialogue information marked with dialogue abstract as training sample;
and training the machine learning model by using the training sample.
5. The method for retrieving a conversation digest of claim 4 wherein,
the step of obtaining the dialogue information marked with the dialogue abstract as a training sample comprises the following steps:
performing similarity matching on each sentence in the dialogue information for training and the artificially generated dialogue abstract to obtain a matching score;
the sentence with the highest matching score is marked as a positive dialogue abstract sample, and other sentences are marked as negative dialogue abstract samples.
6. An apparatus for obtaining a conversation digest, comprising:
the dialogue acquisition module is used for acquiring dialogue information;
the abstract extraction module is used for respectively determining the dialogue abstracts of each dialogue participant from the dialogue information by utilizing a pre-trained machine learning model;
the abstract determining module is used for matching related abstract contents from the dialogue abstracts of all the dialogue participants according to semantic similarity, and generating dialogue abstract combinations as abstracts of the dialogue information;
the abstract determining module is used for combining sentences in dialogue abstracts of different dialogue participators two by two to obtain all candidate abstract combinations; determining whether sentences in each candidate abstract combination are semantically similar; and taking the candidate abstract combination with similar sentence meaning as a dialogue abstract combination.
7. The apparatus for retrieving a conversation digest of claim 6 wherein,
the abstract extraction module is used for dividing the dialogue information into dialogue information of each dialogue participant according to the identification of the dialogue participant; respectively inputting each sentence in the dialogue information of each dialogue participant into a pre-trained machine learning model to determine the probability that each sentence corresponding to the dialogue participant belongs to the abstract sentence; and selecting sentences as dialogue summaries of the dialogue participants according to the probabilities that the sentences corresponding to the dialogue participants belong to the summary sentences.
8. The apparatus for retrieving a conversation digest of claim 6 wherein,
the abstract determining module is used for mapping sentences in the candidate abstract combination into vectors according to a preset rule; and determining whether the two sentences are similar according to the distance between the vectors of the two sentences in the candidate abstract combination.
9. The apparatus for retrieving a conversation digest according to any one of claims 6 to 8, further comprising:
the training sample generation module is used for acquiring dialogue information marked with the dialogue abstract as a training sample;
and the training module is used for training the machine learning model by utilizing the training sample.
10. The apparatus for retrieving a conversation digest of claim 9 wherein,
the training sample generation module is used for carrying out similarity matching on each sentence in the dialogue information for training and the manually generated dialogue abstract to obtain matching scores; the sentence with the highest matching score is marked as a positive dialogue abstract sample, and other sentences are marked as negative dialogue abstract samples.
11. An apparatus for obtaining a conversation digest, comprising:
a memory; and
a processor coupled to the memory, the processor configured to perform the method of retrieving a conversation digest as claimed in any one of claims 1 to 5 based on instructions stored in the memory.
12. A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor realizes the steps of the method according to any of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811483302.3A CN111368066B (en) | 2018-12-06 | 2018-12-06 | Method, apparatus and computer readable storage medium for obtaining dialogue abstract |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811483302.3A CN111368066B (en) | 2018-12-06 | 2018-12-06 | Method, apparatus and computer readable storage medium for obtaining dialogue abstract |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111368066A CN111368066A (en) | 2020-07-03 |
| CN111368066B true CN111368066B (en) | 2024-02-09 |
Family
ID=71205905
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811483302.3A Active CN111368066B (en) | 2018-12-06 | 2018-12-06 | Method, apparatus and computer readable storage medium for obtaining dialogue abstract |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111368066B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP4064110A1 (en) * | 2021-03-24 | 2022-09-28 | Samsung SDS Co., Ltd. | Apparatus and method for training dialogue summary model |
| CN114357987B (en) * | 2021-12-28 | 2024-12-03 | 讯飞智元信息科技有限公司 | Abstract generation method and related device, equipment and storage medium |
| CN114490976B (en) * | 2021-12-30 | 2023-04-25 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for generating dialogue abstract training data |
| CN116933801A (en) * | 2022-04-01 | 2023-10-24 | 北京沃东天骏信息技术有限公司 | Dialogue abstract generation method and device, model training method and device |
| CN117216044A (en) * | 2023-08-31 | 2023-12-12 | 海信集团控股股份有限公司 | Report generation method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1828609A (en) * | 2006-04-13 | 2006-09-06 | 北大方正集团有限公司 | A Topic-Oriented Query-Oriented Multi-Document Summarization Method Based on Cluster Permutation |
| JP2011028638A (en) * | 2009-07-28 | 2011-02-10 | Nippon Telegr & Teleph Corp <Ntt> | Summary sentence creation apparatus, summary sentence creation method and program |
| CN106407178A (en) * | 2016-08-25 | 2017-02-15 | 中国科学院计算技术研究所 | Session abstract generation method and device |
| CN108388597A (en) * | 2018-02-01 | 2018-08-10 | 深圳市鹰硕技术有限公司 | Conference summary generation method and device |
-
2018
- 2018-12-06 CN CN201811483302.3A patent/CN111368066B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1828609A (en) * | 2006-04-13 | 2006-09-06 | 北大方正集团有限公司 | A Topic-Oriented Query-Oriented Multi-Document Summarization Method Based on Cluster Permutation |
| JP2011028638A (en) * | 2009-07-28 | 2011-02-10 | Nippon Telegr & Teleph Corp <Ntt> | Summary sentence creation apparatus, summary sentence creation method and program |
| CN106407178A (en) * | 2016-08-25 | 2017-02-15 | 中国科学院计算技术研究所 | Session abstract generation method and device |
| CN108388597A (en) * | 2018-02-01 | 2018-08-10 | 深圳市鹰硕技术有限公司 | Conference summary generation method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111368066A (en) | 2020-07-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111368066B (en) | Method, apparatus and computer readable storage medium for obtaining dialogue abstract | |
| CN106997375B (en) | Customer service reply recommendation method based on deep learning | |
| CN113313022B (en) | Training method of character recognition model and method for recognizing characters in image | |
| CN111858843B (en) | Text classification method and device | |
| CN111966800B (en) | Emotion dialogue generation method and device and emotion dialogue model training method and device | |
| CN112528637B (en) | Text processing model training method, device, computer equipment and storage medium | |
| CN109376222B (en) | Question-answer matching degree calculation method, question-answer automatic matching method and device | |
| CN111626362B (en) | Image processing method, device, computer equipment and storage medium | |
| CN112164391A (en) | Statement processing method and device, electronic equipment and storage medium | |
| CN107844481B (en) | Recognition text error detection method and device | |
| CN108228576B (en) | Text translation method and device | |
| CN113836303A (en) | A text category identification method, device, computer equipment and medium | |
| CN113094478B (en) | Expression reply method, device, equipment and storage medium | |
| CN114528851B (en) | Reply sentence determination method, reply sentence determination device, electronic equipment and storage medium | |
| CN110532562B (en) | Neural network training method, idiom misuse detection method and device and electronic equipment | |
| CN115272540A (en) | Processing method and device based on virtual customer service image, equipment and medium | |
| CN112349294A (en) | Voice processing method and device, computer readable medium and electronic equipment | |
| CN113010664A (en) | Data processing method and device and computer equipment | |
| CN113705207A (en) | Grammar error recognition method and device | |
| CN115983283A (en) | Emotion classification method and device based on artificial intelligence, computer equipment and medium | |
| CN117216206A (en) | Session processing method and device, electronic equipment and storage medium | |
| CN115906873A (en) | Dialogue processing method and device, and training method and device | |
| CN118132687A (en) | Sentence processing and category model training method, sentence processing and category model training device, sentence processing equipment and category model training medium | |
| CN115630652A (en) | Customer service session emotion analysis system, method and computer system | |
| CN116467414B (en) | Data verification method, device, equipment and computer readable storage medium |
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 |