Information processing method and device, equipment and storage medium

Abstract

The present disclosure provides an information processing method, device and equipment and storage medium, and relates to the field of artificial intelligence customer service. In some embodiments of the present disclosure, a plurality of candidate dialogues are obtained; the plurality of candidate dialogues are merged according to the association relationship between the question texts of adjacent candidate dialogues, and a plurality of merged dialogues are obtained; the plurality of merged dialogues are clustered, and at least one class cluster dialogue group is obtained; a dialogue that meets a popular degree condition is selected from each class cluster dialogue group to form a popular dialogue; a target dialogue that meets a similarity condition is selected from the popular dialogue according to the similarity between the reply text of each popular dialogue and the machine text corresponding to each popular dialogue, and the target dialogue is stored in a dialogue knowledge base; the present disclosure automatically determines the target dialogue from the candidate dialogue to fill the dialogue knowledge base, which is low in labor cost and improves the acquisition efficiency of dialogue knowledge in the dialogue knowledge base.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence customer service, and particularly relates to an information processing method and device, equipment and a storage medium. BACKGROUND

[0002] The consumer electronics industry is a highly competitive and rapidly changing industry, and customer needs and expectations are also constantly improving. In such an environment, the role of customer service is crucial, as it greatly influences customer satisfaction and loyalty. Intelligent customer service systems are systems that use artificial intelligence technology to automatically interact with customers. Compared with traditional manual customer service, intelligent customer service systems can quickly respond to customer needs, provide accurate and effective information and solutions, avoid customer dissatisfaction with waiting and queuing, and increase customer trust and loyalty. At the same time, it greatly saves customer service manpower and achieves the purpose of reducing costs and increasing efficiency for enterprises.

[0003] The customer service knowledge base is the core of the intelligent customer service system, which contains the knowledge related to the products, services, and policies of the enterprise, and is used to answer customer inquiries. Knowledge mining is a process that uses data mining, natural language processing, machine learning, and other technologies to extract valuable knowledge from large amounts of text or voice data. Knowledge mining can extract and organize structured or unstructured knowledge from enterprise business documents, customer service logs, user feedback, and other data sources to form a customer service knowledge base. On the other hand, knowledge mining can also update and optimize the customer service knowledge base by analyzing customer questions and satisfaction, improving its accuracy and coverage, and further improving the dialogue interaction capabilities of intelligent customer service.

[0004] Currently, knowledge mining for customer knowledge bases requires human intervention, which is costly and inefficient in obtaining dialogue knowledge. SUMMARY

[0005] The present disclosure provides an information processing method, device, equipment and storage medium to at least solve the problem of high labor costs and low efficiency in obtaining dialogue knowledge.

[0006] The technical solutions of the present disclosure are as follows:

[0007] The present disclosure provides an information processing method, comprising:

[0008] Obtaining a plurality of candidate dialogues;

[0009] According to the association relationship between the question texts of adjacent candidate dialogues, the plurality of candidate dialogues are merged to obtain a plurality of merged dialogues;

[0010] The plurality of merged dialogues are clustered to obtain at least one class cluster dialogue group;

[0011] selecting a dialogue meeting a popular degree condition from each of the cluster dialogue groups to form a popular dialogue;

[0012] selecting a target dialogue meeting a similarity condition from the popular dialogues according to a similarity between a reply text of each of the popular dialogues and machine text corresponding to each of the popular dialogues, and storing the target dialogue into a dialogue knowledge base.

[0013] Optionally, the obtaining a plurality of candidate dialogues comprises:

[0014] inputting the question text, the entity information, the syntax feature and the question text of the historical dialogue of each original dialogue into the trained classification model to obtain an intent category of each original dialogue;

[0015] selecting a plurality of candidate dialogues with the intent category being valid question answering from the plurality of original dialogues.

[0016] Optionally, the classification model comprises an embedding layer, a deep neural network layer and an activation function layer, and the inputting the entity information, the syntax feature and the historical dialogue of each original dialogue into the trained classification model to obtain the intent category of each original dialogue comprises:

[0017] inputting the question text of each original dialogue and the question text of the historical dialogue of each original dialogue into the embedding layer to obtain first dialogue features;

[0018] inputting the entity information and the syntax feature of each original dialogue into the deep neural network layer to obtain second dialogue features;

[0019] inputting the first dialogue features and the second dialogue features into the activation function layer to obtain the intent category of each original dialogue.

[0020] Optionally, the merging a plurality of candidate dialogues according to an association relationship between question texts of adjacent candidate dialogues to obtain a plurality of merged dialogues comprises:

[0021] inputting question texts of two adjacent candidate dialogues into an embedding layer to obtain a first embedding feature vector and a second embedding feature vector;

[0022] calculating a first similarity of the first embedding feature vector and the second embedding feature vector;

[0023] in a case where the first similarity is greater than or equal to a first similarity threshold, merging the two adjacent candidate dialogues;

[0024] The question texts of the subsequent adjacent candidate dialogues are sequentially merged to obtain a plurality of merged dialogues.

[0025] Optionally, the plurality of merged dialogues are clustered to obtain at least one cluster dialogue group, including:

[0026] The question text and the reply text of each merged dialogue are spliced to obtain spliced text of each merged dialogue;

[0027] The spliced text of each merged dialogue is input into an embedding layer to obtain a third embedding feature vector of each merged dialogue;

[0028] The third embedding feature vectors of the plurality of merged dialogues are clustered to obtain a cluster to which each merged dialogue belongs;

[0029] According to the cluster to which each merged dialogue belongs, at least one cluster dialogue group is obtained.

[0030] Optionally, a dialogue satisfying a hot degree condition is selected from each cluster dialogue group to form a hot dialogue, including:

[0031] For a target cluster dialogue group, a hot degree value of each dialogue in the target cluster dialogue group is calculated according to a frequency of a similar question corresponding to each dialogue in the target cluster dialogue group, a negative feedback rate of the similar question, and a cluster center distance, wherein the target cluster dialogue group is any one of the cluster dialogue groups;

[0032] The dialogue with the highest hot degree value is selected from the target cluster dialogue group as a hot dialogue.

[0033] Optionally, a target dialogue satisfying a similarity condition is selected from the hot dialogue according to a similarity between the reply text of each hot dialogue and machine text corresponding to each hot dialogue, including:

[0034] Machine text corresponding to the question text of each hot dialogue is obtained;

[0035] The reply text of each hot dialogue and the machine text corresponding to each hot dialogue are input into an embedding layer to obtain a fourth embedding feature vector and a fifth embedding feature vector;

[0036] A second similarity between the fourth embedding feature vector and the fifth embedding feature vector is calculated;

[0037] A target dialogue with the second similarity less than a second similarity threshold is selected from the hot dialogue.

[0038] The embodiments of the present disclosure further provide an information processing device, comprising:

[0039] an acquisition module, configured to acquire a plurality of candidate dialogues;

[0040] a merging module, configured to merge the plurality of candidate dialogues according to the association relationship between the question texts of adjacent candidate dialogues, to obtain a plurality of merged dialogues;

[0041] a clustering module, configured to cluster the plurality of merged dialogues, to obtain at least one cluster dialogue group;

[0042] a first selection module, configured to select a dialogue that meets a hot degree condition from each cluster dialogue group, to form a hot dialogue;

[0043] a second selection module, configured to select a target dialogue that meets a similarity condition from the hot dialogue according to the similarity between the reply text of each hot dialogue and the machine text corresponding to each hot dialogue, and store the target dialogue into a dialogue knowledge base.

[0044] Optionally, the acquisition module, when acquiring the plurality of candidate dialogues, is configured to:

[0045] input the question text, entity information, syntactic feature and question text of the historical dialogue of each original dialogue into the trained classification model, to obtain the intent category of each original dialogue;

[0046] select a plurality of candidate dialogues with the intent category being valid question answering from the plurality of original dialogues.

[0047] Optionally, the classification model comprises an embedding layer, a deep neural network layer and an activation function layer, and the acquisition module, when inputting the entity information, syntactic feature and historical dialogue of each original dialogue into the trained classification model to obtain the intent category of each original dialogue, is configured to:

[0048] input the question text of each original dialogue and the question text of the historical dialogue of each original dialogue into the embedding layer, to obtain first dialogue features;

[0049] input the entity information and syntactic feature of each original dialogue into the deep neural network layer, to obtain second dialogue features;

[0050] input the first dialogue features and the second dialogue features into the activation function layer, to obtain the intent category of each original dialogue.

[0051] Optionally, the merging module, when merging the plurality of candidate dialogues according to the association relationship between the question texts of adjacent candidate dialogues, to obtain the plurality of merged dialogues, is configured to:

[0052] inputting the question text of the two adjacent candidate dialogues into an embedding layer to obtain a first embedding feature vector and a second embedding feature vector;

[0053] calculating a first similarity of the first embedding feature vector and the second embedding feature vector;

[0054] in a case where the first similarity is greater than or equal to a first similarity threshold, merging the two adjacent candidate dialogues;

[0055] sequentially merging the question text of the subsequent adjacent candidate dialogues to obtain a plurality of merged dialogues.

[0056] Optionally, when the clustering module clusters the plurality of merged dialogues to obtain at least one cluster dialogue group, the clustering module is configured to:

[0057] splicing the question text and the reply text of each merged dialogue to obtain spliced text of each merged dialogue;

[0058] inputting the spliced text of each merged dialogue into an embedding layer to obtain a third embedding feature vector of each merged dialogue;

[0059] clustering the third embedding feature vectors of the plurality of merged dialogues to obtain a cluster to which each merged dialogue belongs;

[0060] grouping the clusters to which each merged dialogue belongs to obtain at least one cluster dialogue group.

[0061] Optionally, when the first selection module selects a dialogue that meets a hot degree condition from each cluster dialogue group to form a hot dialogue, the first selection module is configured to:

[0062] for a target cluster dialogue group, calculating a hot degree value of each dialogue in the target cluster dialogue group according to a frequency of a similar question corresponding to the dialogue, a negative feedback rate of the similar question, and a cluster center distance, wherein the target cluster dialogue group is any one of the cluster dialogue groups;

[0063] selecting a dialogue with the highest hot degree value from the target cluster dialogue group as a hot dialogue.

[0064] Optionally, when the second selection module selects a target dialogue that meets a similarity condition from the hot dialogues according to a similarity between the reply text of each hot dialogue and machine text corresponding to the hot dialogue, the second selection module is configured to:

[0065] obtain machine text corresponding to the question text of each of the hot conversations;

[0066] input the reply text of each of the hot conversations and the machine text corresponding to each of the hot conversations into an embedding layer to obtain a fourth embedding feature vector and a fifth embedding feature vector;

[0067] calculate a second similarity between the fourth embedding feature vector and the fifth embedding feature vector;

[0068] select a target conversation from the hot conversations, where the second similarity of the target conversation is less than a second similarity threshold.

[0069] The embodiments of the present disclosure further provide an electronic device, comprising:

[0070] a processor;

[0071] a memory for storing instructions executable by the processor;

[0072] The processor is configured to execute the instructions to implement the steps in the above method.

[0073] The embodiments of the present disclosure further provide a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps in the above method.

[0074] The embodiments of the present disclosure further provide a computer program product comprising computer programs / instructions, the computer programs / instructions being executed by a processor to implement the steps in the above method.

[0075] The embodiments of the present disclosure provide at least the following beneficial effects:

[0076] In some embodiments of the present disclosure, a plurality of candidate conversations are obtained; the plurality of candidate conversations are merged according to the association relationship between the question texts of adjacent candidate conversations to obtain a plurality of merged conversations; the plurality of merged conversations are clustered to obtain at least one cluster conversation group; a conversation meeting a hot degree condition is selected from each cluster conversation group to form a hot conversation; a target conversation meeting a similarity condition is selected from the hot conversations according to the similarity between the reply text of each hot conversation and the machine text corresponding to each hot conversation, and the target conversation is stored in a conversation knowledge base; the present disclosure automatically determines the target conversation from the candidate conversations to fill the conversation knowledge base, which is low in labor cost and improves the acquisition efficiency of conversation knowledge in the conversation knowledge base.

[0077] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0078] The drawings incorporated into the specification and forming a part thereof illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure, and are not intended to represent the only embodiments consistent with the present disclosure.

[0079] Figure 1 A flowchart of an information processing method provided for an exemplary embodiment of the present disclosure;

[0080] Figure 2 A flowchart of original dialogue intent recognition provided for an embodiment of the present disclosure;

[0081] Figure 3 A flowchart of dialogue merging provided for an exemplary embodiment of the present disclosure;

[0082] Figure 4 A flowchart of screening target dialogue provided for an embodiment of the present disclosure;

[0083] Figure 5 A structural schematic diagram of an information processing device provided for an exemplary embodiment of the present disclosure;

[0084] Figure 6 A structural schematic diagram of an electronic device provided for an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0085] In order to make the ordinary person skilled in the art better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below in conjunction with the drawings.

[0086] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation described in the following exemplary embodiments does not represent all implementations consistent with the present disclosure. Rather, they are only examples of devices and methods consistent with some aspects of the present disclosure.

[0087] It should be noted that the user information involved in the present disclosure includes but is not limited to user equipment information and user personal information; the collection, storage, use, processing, transmission, provision and disclosure of user information in the present disclosure comply with the provisions of relevant laws and regulations, and do not violate public order and good customs.

[0088] Currently, the knowledge mining in the field of intelligent customer service is mainly based on user negative feedback data and some artificially labeled negative samples. Among them, user negative feedback data refers to negative comments and feedback information given by users during product use, such as complaints, returns, customer service inquiries, etc. Negative feedback data reflects the satisfaction and dissatisfaction of users with the product, and is of great significance for product improvement and user experience improvement. The algorithms involved mainly include the following:

[0089] Association rule mining algorithms, such as Apriori algorithm, FP-Growth algorithm, etc. Mainly through the mining of frequently occurring keywords, entities, intents, etc. in the problem text.

[0090] Statistical correlation algorithms, such as TF-IDF, LDA, etc. This kind of algorithm extracts and filters information by counting the word frequency, co-occurrence frequency, information gain, etc. in the dialogue.

[0091] Neural network-based algorithms, which directly associate the input and output in the dialogue by designing different neural network structures to improve the expression ability and generalization ability of the model, such as using CNN, RNN, Transformer, etc. Model to generate replies, summaries, etc. in the dialogue.

[0092] However, the mining scheme based on negative feedback data has the following problems: User negative feedback data is often very scarce, and it is difficult to directly know the user's true intention from user feedback. The sample efficiency obtained from artificial labeling is very low, and the labeling cost is relatively high, making it difficult to form a continuous optimization process. User negative feedback data only contains the user's problem, and the solution to the problem is missing, which requires a lot of manual work to continue knowledge mining, increasing the cost of manual work.

[0093] In order to solve the above technical problems, in some embodiments of the present disclosure, a plurality of candidate dialogues are obtained; the plurality of candidate dialogues are merged according to the association relationship between the problem texts of adjacent candidate dialogues, to obtain a plurality of merged dialogues; the plurality of merged dialogues are clustered to obtain at least one class cluster dialogue group; a dialogue that meets the hot degree condition is selected from each class cluster dialogue group to form a hot dialogue; a target dialogue that meets the similarity condition is selected from the hot dialogue according to the similarity between the reply text of each hot dialogue and the machine text corresponding to each hot dialogue, and the target dialogue is stored in the dialogue knowledge base; The present disclosure automatically determines the target dialogue from the candidate dialogue to fill the dialogue knowledge base, which has a lower labor cost and improves the efficiency of obtaining dialogue knowledge in the dialogue knowledge base.

[0094] The technical solutions provided by the embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0095] Figure 1A flowchart of an information processing method is provided for an exemplary embodiment of the present disclosure. As shown in Figure 1 The method includes:

[0096] S101: obtaining a plurality of candidate dialogues;

[0097] S102: merging the plurality of candidate dialogues according to the association relationship between the question texts of adjacent candidate dialogues, to obtain a plurality of merged dialogues;

[0098] S103: clustering the plurality of merged dialogues to obtain at least one cluster dialogue group;

[0099] S104: selecting a dialogue that meets a hot degree condition from each cluster dialogue group to form a hot dialogue;

[0100] S105: selecting a target dialogue that meets a similarity condition from the hot dialogue according to the similarity between the reply text of each hot dialogue and the machine text corresponding to each hot dialogue, and storing the target dialogue into a dialogue knowledge base.

[0101] In the present embodiment, the execution subject of the above method can be an intelligent terminal or a server.

[0102] In the present embodiment, the type of the intelligent terminal is not limited, including but not limited to the following: personal computer, notebook computer, mobile phone, smart bracelet, smart watch and smart sound.

[0103] In the present embodiment, the implementation form of the server. For example, the server can be a conventional server, a cloud server, a cloud host, a virtual center, etc. The server mainly includes a processor, a hard disk, a memory, a system bus, etc., and a general computer architecture type.

[0104] In the present embodiment, a plurality of candidate dialogues are obtained; the plurality of candidate dialogues are merged according to the association relationship between the question texts of adjacent candidate dialogues, to obtain a plurality of merged dialogues; the plurality of merged dialogues are clustered to obtain at least one cluster dialogue group; a dialogue that meets a hot degree condition is selected from each cluster dialogue group to form a hot dialogue; a target dialogue that meets a similarity condition is selected from the hot dialogue according to the similarity between the reply text of each hot dialogue and the machine text corresponding to each hot dialogue, and the target dialogue is stored into a dialogue knowledge base; the present disclosure automatically determines the target dialogue from the candidate dialogue to fill the dialogue knowledge base, which is low in labor cost and improves the acquisition efficiency of dialogue knowledge in the dialogue knowledge base.

[0105] In some embodiments of the present disclosure, a plurality of candidate dialogues is obtained. One implementable manner is to input the question text, entity information, syntactic features and question text of the historical dialogue of each original dialogue into a classification model that has been trained to obtain the intent category of each original dialogue. The intent category of the original dialogue includes but is not limited to casual chat, invalid question text, ambiguous intent and valid question. The embodiments of the present disclosure classify the intent of the original dialogue by using the classification model, thereby improving the efficiency of intent recognition and improving the accuracy of intent recognition.

[0106] In the above embodiments, the classification model includes an embedding layer, a deep neural network layer and an activation function layer. The entity information, syntactic features and historical dialogue of each original dialogue are input into the classification model that has been trained to obtain the intent category of each original dialogue. One implementable manner is to input the question text of each original dialogue and the question text of the historical dialogue of each original dialogue into the embedding layer to obtain first dialogue features. The entity information and syntactic features of each original dialogue are input into the deep neural network layer to obtain second dialogue features. The first dialogue features and the second dialogue features are input into the activation function layer to obtain the intent category of each original dialogue. It should be noted that the embedding layer in the embodiments of the present disclosure can be the embedding layer of the pre-trained RoBERTa model, so that the classification model has strong generalization ability. In combination with the question text of each original dialogue and the question text of the historical dialogue of each original dialogue, the classification model can infer the intent of the user in combination with the context, thereby further improving the accuracy of intent recognition. The entity information and syntactic features are added as additional features, so that the classification model has stronger recognition ability for invalid intent, inverted sentences, misspelled words and the like.

[0107] An exemplary Figure 2 A flowchart of an original dialogue intent recognition process is provided in the embodiments of the present disclosure. As shown in Figure 2 , any one original dialogue includes a user query and a customer answer. The question text (original query) of any one original dialogue, entity information, syntactic features and question text (historical query) of the historical dialogue are input into the RoBERTa model and the DNN deep neural network layer to obtain the intent category (Query category) corresponding to the original dialogue. Finally, the dialogues with intent categories such as casual chat, invalid question text (invalid query) and ambiguous intent are deleted, and the dialogues with the intent category of valid question are retained as candidate dialogues.

[0108] In some embodiments of the present disclosure, the plurality of candidate dialogues are merged according to the association between the question texts of adjacent candidate dialogues, to obtain a plurality of merged dialogues. One implementable manner is to input the question texts of the first two adjacent candidate dialogues into an embedding layer to obtain a first embedding feature vector and a second embedding feature vector; calculate a first similarity of the first embedding feature vector and the second embedding feature vector; in the case where the first similarity is greater than or equal to a first similarity threshold, merge any two adjacent candidate dialogues; and sequentially merge the question texts of the subsequent adjacent candidate dialogues to obtain a plurality of merged dialogues. It should be noted that the first similarity threshold is not limited in the embodiments of the present disclosure, and the first similarity threshold can be adjusted according to actual conditions. Wherein, the continuous user dialogue is cut into a series of question-answer pairs; it is determined whether the intents of the two question texts in a dialogue are strongly associated, and the association of the question texts is determined by using the distance thereof in the semantic space. The embedding layer of the uppermost layer of the pre-trained RoBERTa model can be used to construct the semantic space, and the modeling process of the pre-trained model on a large-scale corpus can ensure stronger representation capability in depicting semantic similarity. The distance between the embedding feature vectors of the two adjacent question texts is used as the standard for determining strong association.

[0109] Exemplarily, Figure 3 A flowchart of dialogue merging is provided for exemplary embodiments of the present disclosure. As shown in Figure 3 Two adjacent candidate dialogues QA include candidate dialogue one and candidate dialogue two. Wherein, the question text of candidate dialogue one is user query1, and the reply text is customer answer1; the question text of candidate dialogue two is user query2, and the reply text is customer answer2; the distance between the two question texts can be represented by similarity. In the case where the first similarity is greater than or equal to the first similarity threshold, any two adjacent candidate dialogues are merged, user query1 and user query2 are merged, and customer answer1 and customer answer2 are merged, and in the case where the first similarity is less than the first similarity threshold, no merging is performed; the QAs in the entire dialogue are iteratively determined, if the QAs are merged, the next round of QA splitting is continued, if the QAs are split, the split data is stored, and a dialogue pair database is constructed. After each QA splicing and merging, the new question text needs to be spliced.

[0110] In some embodiments of the present disclosure, the plurality of merged dialogues are clustered to obtain at least one cluster dialogue group. One implementable manner is to splice the question text and the reply text of each merged dialogue to obtain the spliced text of each merged dialogue; input the spliced text of each merged dialogue into an embedding layer to obtain a third embedding feature vector of each merged dialogue; cluster the third embedding feature vectors of the plurality of merged dialogues to obtain the cluster to which each merged dialogue belongs; and group the clusters to which each merged dialogue belongs to obtain at least one cluster dialogue group. The present disclosure is independent between the segmented merged dialogues; the merged dialogues are classified and the core cluster in the database is extracted. The question text and the reply text of each merged dialogue are spliced to obtain the spliced text of each merged dialogue; the spliced text of each merged dialogue is input into the embedding layer of the RoBERTa model to obtain a third embedding feature vector of each merged dialogue; and the kmeans clustering algorithm is used to obtain the cluster identifier corresponding to each merged dialogue. The present embodiment takes into account the question text and the reply text, and the embedding feature vector can cluster the question texts with similar semantics together, and it is possible that similar question methods of different question texts are also clustered together. For example: both are parameter questions, but different products are inquired. In this case, the semantic discrimination has certain limitations, the semantics of the reply text is also introduced, the distance between the merged texts with the same question method but different answers is enlarged, so that the clustering effect is better and more robust.

[0111] In the above embodiment, the kmeans clustering algorithm can obtain the cluster center vector and the cluster center distance of any dialogue in the cluster dialogue group.

[0112] In some embodiments of the present disclosure, a dialogue that meets the hot degree condition is selected from each cluster dialogue group to form a hot dialogue, that is, the most representative dialogue is selected from each cluster dialogue group. One implementable manner is that, for a target cluster dialogue group, the hot degree value of each dialogue in the target cluster dialogue group is calculated according to the frequency of the similar question corresponding to each dialogue in the target cluster dialogue group, the negative feedback rate of the similar question, and the cluster center distance, wherein the target cluster dialogue group is any dialogue group in the cluster dialogue group; and the dialogue with the highest hot degree value is selected from the target cluster dialogue group as a hot dialogue.

[0113] In the above embodiment, at present, the way of selecting a popular dialogue is to take the sample closest to the cluster center of clustering as the popular dialogue, only considering the similarity in the semantic level, without considering the user preference, the effectiveness of the question and answer, and the like. The embodiment of the present disclosure improves the cluster center distance, introduces the occurrence of similar questions and the corresponding negative feedback in the online question and answer as a correction, extracts the user query method most frequently asked, with the highest negative feedback, and most representative of the cluster. According to the frequency of the occurrence of similar questions corresponding to each dialogue in the target cluster dialogue group, the negative feedback rate of the similar questions, and the cluster center distance, the popular degree value of each dialogue in the target cluster dialogue group is calculated respectively.

[0114] The calculation formula of the popular degree value is as follows:

[0115] Score pair =w1*log(PV simi_query )*(1-nega simi_query )-w2*dist pair

[0116] PV simi_query represents the statistical PV of the online question and answer query similar to the query in the QA pair. Here, the logarithmic function is used to eliminate the excessive contribution of PV when it is too large. nega simi_query represents the negative feedback rate of the online question and answer query similar to the query, and the query with a high negative feedback is punished. dist pair represents the cluster center distance of the QA, and the smaller the distance, the higher the score. w1 and w2 are model hyperparameters, which need to be determined by adjusting the parameters in combination with the business scenario.

[0117] In the above embodiment, a popular dialogue is selected from each cluster dialogue group in turn to form a popular dialogue.

[0118] In some embodiments of the present disclosure, a target dialogue that meets a similarity condition is selected from the popular dialogue according to the similarity between the reply text of each popular dialogue and the machine text corresponding to each popular dialogue. One realizable way is to obtain the machine text corresponding to the question text of each popular dialogue; the reply text of each popular dialogue and the machine text corresponding to each popular dialogue are input into an embedding layer to obtain a fourth embedding feature vector and a fifth embedding feature vector; a second similarity between the fourth embedding feature vector and the fifth embedding feature vector is calculated; and a target dialogue with a second similarity less than a second similarity threshold is selected from the popular dialogue. It should be noted that the second similarity threshold is not limited in the embodiment of the present disclosure, and can be adjusted according to the actual situation.

[0119] Exemplarily, Figure 4 is a flowchart of screening a target dialogue provided by an embodiment of the present disclosure. As shown inFigure 4 As shown, the question text (user query) of any one hot conversation is input into the online intelligent robot to obtain machine text (intelligent customer service robot answer), the reply text (customer answer) of the hot conversation and the machine text corresponding to each hot conversation are input into the embedding layer of the RoBERTa model to obtain the fourth embedding feature vector and the fifth embedding feature vector, and the second similarity of the fourth embedding feature vector and the fifth embedding feature vector is calculated; a target conversation with a second similarity less than a second similarity threshold is selected from the hot conversation, and the target conversation is stored in the conversation knowledge base, directly supplementing the conversation knowledge base and improving the maintenance efficiency of the conversation knowledge base.

[0120] The embodiment of the present disclosure realizes the automatic mining of the conversation between the customer service and the user. The machine learning technology is used to solve the problems of large noise, difficult segmentation, and fuzzy topic extraction in the real conversation process. The maintenance efficiency of the intelligent customer service conversation knowledge base is improved, and the question and answer effect of the intelligent customer service is further improved.

[0121] In the method embodiment of the present disclosure, a plurality of candidate conversations are obtained; the plurality of candidate conversations are merged according to the association relationship between the question texts of adjacent candidate conversations to obtain a plurality of merged conversations; the plurality of merged conversations are clustered to obtain at least one class cluster conversation group; a conversation meeting a hot degree condition is selected from each class cluster conversation group to form a hot conversation; a target conversation meeting a similarity condition is selected from the hot conversation according to the similarity between the reply text of each hot conversation and the machine text corresponding to each hot conversation, and the target conversation is stored in the conversation knowledge base; the target conversation is automatically determined from the candidate conversation to fill the conversation knowledge base, the labor cost is low, and the acquisition efficiency of the conversation knowledge in the conversation knowledge base is improved.

[0122] Figure 5 A structural schematic diagram of an information processing device 50 provided for an exemplary embodiment of the present disclosure is shown. As shown, the information processing device 50 includes an acquisition module 51, a merging module 52, a clustering module 53, a first selection module 54, and a second selection module 55. Figure 5

[0123] The acquisition module 51 is configured to obtain a plurality of candidate conversations.

[0124] The merging module 52 is configured to merge the plurality of candidate conversations according to the association relationship between the question texts of adjacent candidate conversations to obtain a plurality of merged conversations.

[0125] The clustering module 53 is configured to cluster the plurality of merged conversations to obtain at least one class cluster conversation group.

[0126] ​The first selection module 54 is configured to select a dialogue that meets a popular degree condition from each cluster of dialogues to form a popular dialogue.

[0127] The second selection module 55 is configured to select a target dialogue that meets a similarity condition from the popular dialogues according to a similarity between a reply text of each popular dialogue and machine text corresponding to each popular dialogue, and store the target dialogue into a dialogue knowledge base.

[0128] Optionally, the obtaining module 51 is configured to:

[0129] input the question text, the entity information, the syntactic feature and the question text of the historical dialogue of each original dialogue into the trained classification model to obtain the intent category of each original dialogue.

[0130] select a plurality of candidate dialogues with the intent category of effective question answering from the plurality of original dialogues.

[0131] Optionally, the classification model comprises an embedding layer, a deep neural network layer and an activation function layer, and the obtaining module 51 is configured to:

[0132] input the question text of each original dialogue and the question text of the historical dialogue of each original dialogue into the embedding layer to obtain first dialogue features;

[0133] input the entity information and the syntactic feature of each original dialogue into the deep neural network layer to obtain second dialogue features;

[0134] input the first dialogue features and the second dialogue features into the activation function layer to obtain the intent category of each original dialogue.

[0135] Optionally, the merging module 52 is configured to:

[0136] input the question texts of the first two adjacent candidate dialogues into the embedding layer to obtain a first embedding feature vector and a second embedding feature vector;

[0137] calculate a first similarity of the first embedding feature vector and the second embedding feature vector;

[0138] merge any two adjacent candidate dialogues in a case where the first similarity is greater than or equal to a first similarity threshold;

[0139] merge the question texts of the subsequent adjacent candidate dialogues in sequence to obtain a plurality of merged dialogues.

[0140] Optionally, the clustering module 53, when clustering the plurality of merged dialogues to obtain at least one cluster dialogue group, is configured to:

[0141] concatenate the question text and the reply text of each merged dialogue to obtain concatenated text of each merged dialogue;

[0142] input the concatenated text of each merged dialogue into the embedding layer to obtain a third embedding feature vector of each merged dialogue;

[0143] cluster the third embedding feature vectors of the plurality of merged dialogues to obtain a cluster to which each merged dialogue belongs;

[0144] group the clusters to which each merged dialogue belongs to obtain at least one cluster dialogue group.

[0145] Optionally, the first selection module 54, when selecting a dialogue that meets a hot degree condition from each cluster dialogue group to form a hot dialogue, is configured to:

[0146] for a target cluster dialogue group, calculate a hot degree value of each dialogue in the target cluster dialogue group according to a frequency of a similar question corresponding to each dialogue in the target cluster dialogue group, a negative feedback rate of the similar question, and a cluster center distance, wherein the target cluster dialogue group is any one of the cluster dialogue groups;

[0147] select a dialogue with the highest hot degree value from the target cluster dialogue group as a hot dialogue.

[0148] Optionally, the second selection module 55, when selecting a target dialogue that meets a similarity condition from the hot dialogues according to a similarity between the reply text of each hot dialogue and machine text corresponding to each hot dialogue, is configured to:

[0149] obtain machine text corresponding to the question text of each hot dialogue;

[0150] input the reply text of each hot dialogue and the machine text corresponding to each hot dialogue into the embedding layer to obtain a fourth embedding feature vector and a fifth embedding feature vector;

[0151] calculate a second similarity between the fourth embedding feature vector and the fifth embedding feature vector;

[0152] select a target dialogue from the hot dialogues, the second similarity of which is less than a second similarity threshold.

[0153] As to the apparatus in the above-described embodiments, the specific manners in which various modules perform operations have been described in details in the embodiments of the method, and thus will not be described in details here.

[0154] Figure 6A structural schematic diagram of an electronic device is provided for an exemplary embodiment of the present disclosure. As shown in Figure 6 The electronic device includes a memory 61 and a processor 62. In addition, the electronic device also includes a power supply component 63 and a communication component 64.

[0155] The memory 61 is configured to store computer programs and can be configured to store other various data to support operations on the electronic device. Examples of the data include instructions for any application program or method operating on the electronic device.

[0156] The memory 61 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0157] The communication component 64 is configured to perform data transmission with other devices.

[0158] The processor 62 can execute computer instructions stored in the memory 61 to: obtain a plurality of candidate dialogues;

[0159] According to the association relationship between the question texts of adjacent candidate dialogues, the plurality of candidate dialogues are merged to obtain a plurality of merged dialogues;

[0160] The plurality of merged dialogues are clustered to obtain at least one cluster dialogue group;

[0161] A dialogue satisfying a hot degree condition is selected from each cluster dialogue group to form a hot dialogue;

[0162] According to the similarity between the reply text of each hot dialogue and the machine text corresponding to each hot dialogue, a target dialogue satisfying a similarity condition is selected from the hot dialogue, and the target dialogue is stored in a dialogue knowledge base.

[0163] Optionally, the processor 62, when obtaining the plurality of candidate dialogues, is configured to:

[0164] The question text, entity information, syntactic feature and question text of the historical dialogue of each original dialogue are input into a trained classification model to obtain an intent category of each original dialogue;

[0165] A plurality of candidate dialogues with the intent category being effective question answering are selected from the plurality of original dialogues.

[0166] Optionally, the classification model comprises an embedding layer, a deep neural network layer and an activation function layer, and the processor 62 is configured to:

[0167] input the question text of each original dialogue and the question text of the historical dialogue of each original dialogue into the embedding layer to obtain first dialogue features;

[0168] input the entity information and the syntax features of each original dialogue into the deep neural network layer to obtain second dialogue features;

[0169] input the first dialogue features and the second dialogue features into the activation function layer to obtain the intent category of each original dialogue.

[0170] Optionally, the processor 62 is configured to:

[0171] input the question text of the first two adjacent candidate dialogues into the embedding layer to obtain a first embedding feature vector and a second embedding feature vector;

[0172] calculate a first similarity between the first embedding feature vector and the second embedding feature vector;

[0173] merge any two adjacent candidate dialogues if the first similarity is greater than or equal to a first similarity threshold;

[0174] merge the question texts of the subsequent adjacent candidate dialogues in sequence to obtain a plurality of merged dialogues.

[0175] Optionally, the processor 62 is configured to:

[0176] splice the question text and the reply text of each merged dialogue to obtain spliced text of each merged dialogue;

[0177] input the spliced text of each merged dialogue into the embedding layer to obtain a third embedding feature vector of each merged dialogue;

[0178] cluster the third embedding feature vectors of the plurality of merged dialogues to obtain a cluster to which each merged dialogue belongs;

[0179] group the clusters according to the cluster to which each merged dialogue belongs to obtain at least one cluster dialogue group.

[0180] Optionally, the processor 62 is configured to:

[0181] According to the frequency of the similar question corresponding to each dialogue in the target class cluster dialogue group, the negative feedback rate of the similar question, and the class cluster center distance, a popular degree value of each dialogue in the target class cluster dialogue group is calculated, wherein the target class cluster dialogue group is any one of the class cluster dialogue groups.

[0182] A dialogue with the highest popular degree value is selected from the target class cluster dialogue group as a popular dialogue.

[0183] Optionally, when the processor 62 selects a target dialogue that meets a similarity condition from the popular dialogues according to the similarity between the reply text of each popular dialogue and the machine text corresponding to each popular dialogue, the processor 62 is configured to:

[0184] Obtain the machine text corresponding to the question text of each popular dialogue.

[0185] The reply text of each popular dialogue and the machine text corresponding to each popular dialogue are input into an embedding layer to obtain a fourth embedding feature vector and a fifth embedding feature vector.

[0186] Calculate a second similarity between the fourth embedding feature vector and the fifth embedding feature vector.

[0187] Select a target dialogue with a second similarity less than a second similarity threshold from the popular dialogues.

[0188] Correspondingly, the embodiments of the present disclosure also provide a computer readable storage medium storing a computer program. When the computer readable storage medium stores the computer program, and the computer program is executed by one or more processors, the one or more processors are caused to perform Figure 1 the steps in the method embodiments.

[0189] Correspondingly, the embodiments of the present disclosure also provide a computer program product, which includes a computer program / instruction. When the computer program / instruction is executed by a processor Figure 1 the steps in the method embodiments.

[0190] The above Figure 6The communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA) technology, Ultra-Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0191] The above Figure 6 The power supply component provides power to the various components of the device in which it resides. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which it resides.

[0192] Electronic devices also include displays and audio components.

[0193] The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0194] An audio component may be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals may be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0195] In the device, the equipment and the computer program product embodiments of the present disclosure, a plurality of candidate dialogues are acquired; the plurality of candidate dialogues are merged according to the association relationship between the question texts of adjacent candidate dialogues, and a plurality of merged dialogues are obtained; the plurality of merged dialogues are clustered, and at least one cluster dialogue group is obtained; a dialogue meeting a popular degree condition is selected from each cluster dialogue group to form a popular dialogue; a target dialogue meeting a similarity condition is selected from the popular dialogue according to the similarity between the reply text of each popular dialogue and the machine text corresponding to each popular dialogue, and the target dialogue is stored in a dialogue knowledge base; the present disclosure automatically determines the target dialogue from the candidate dialogue to fill the dialogue knowledge base, and the labor cost is low, and the acquisition efficiency of dialogue knowledge in the dialogue knowledge base is improved.

[0196] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Accordingly, the present application can be embodied in the form of complete hardware embodiments, complete software embodiments, or embodiments combining software and hardware aspects. Moreover, the present application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) having computer-usable program code embodied thereon.

[0197] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.

[0198] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that implements the functions specified in the flowcharts and / or block diagrams.

[0199] These computer program instructions can also be loaded into 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 block or blocks. Figure 1 Figure 1

[0200] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0201] The memory can include non-persistent memory and / or persistent memory, such as flash memory, read-only memory (ROM), or other memory. The memory is an example of computer readable media.

[0202] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0203] ​​It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0204] The above are merely specific embodiments of this disclosure, enabling those skilled in the art to understand or implement this disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An information processing method, characterized in that, include: The intent of the original dialogue is classified using a pre-trained classification model to obtain the intent category of each original dialogue, and multiple candidate dialogues with the intent category being valid question answering are obtained from multiple original dialogues. Based on the correlation between the question texts of adjacent candidate dialogues, multiple candidate dialogues are merged to obtain multiple merged dialogues; wherein, the question texts of the first two adjacent candidate dialogues are input into the embedding layer to obtain a first embedding feature vector and a second embedding feature vector; a first similarity between the first embedding feature vector and the second embedding feature vector is calculated; if the first similarity is greater than or equal to a first similarity threshold, any two adjacent candidate dialogues are merged; the question texts of subsequent adjacent candidate dialogues are merged in turn to obtain multiple merged dialogues; Cluster the multiple merged dialogues to obtain at least one cluster of dialogue groups; For a target cluster of dialogue groups, based on the frequency of similar questions, the negative feedback rate of similar questions, and the cluster center distance for each dialogue in the target cluster of dialogue groups, the popularity value of each dialogue in the target cluster of dialogue groups is calculated. The dialogue with the highest popularity value is selected from the target cluster of dialogue groups as a popular dialogue. Wherein, the target cluster of dialogue groups can be any dialogue group in the cluster of dialogue groups. Based on the similarity between the response text of each popular dialogue and the machine text corresponding to each popular dialogue, target dialogues that meet the similarity criteria are selected from the popular dialogues and stored in the dialogue knowledge base.

2. The method according to claim 1, characterized in that, The step of classifying the intent of the original dialogue using a pre-trained classification model to obtain the intent category of each original dialogue includes: Input the question text, entity information, syntactic features, and question text of historical dialogues for each original dialogue into the trained classification model to obtain the intent category of each original dialogue.

3. The method according to claim 2, characterized in that, The classification model includes an embedding layer, a deep neural network layer, and an activation function layer. The entity information, syntactic features, and historical dialogues of each original dialogue are input into the trained classification model to obtain the intent category of each original dialogue, including: The question text of each original dialogue and the question text of each historical dialogue of the original dialogue are input into the embedding layer to obtain the first dialogue feature; The entity information and syntactic features of each original dialogue are input into a deep neural network layer to obtain the second dialogue features; The first dialogue feature and the second dialogue feature are input into the activation function layer to obtain the intent category of each original dialogue.

4. The method according to claim 1, characterized in that, The process of clustering multiple merged dialogues to obtain at least one group of dialogue clusters includes: The question text and response text of each merged dialogue are concatenated to obtain the concatenated text of each merged dialogue; The concatenated text of each merged dialogue is input into the embedding layer to obtain the third embedding feature vector of each merged dialogue; Cluster the third embedding feature vectors of multiple merged dialogues to obtain the cluster to which each merged dialogue belongs; Grouping is performed based on the cluster to which each merged dialogue belongs, resulting in at least one clustered dialogue group.

5. The method according to claim 1, characterized in that, The step of selecting target dialogues that meet the similarity criteria from the popular dialogues based on the similarity between the response text of each popular dialogue and the machine text corresponding to each popular dialogue includes: Obtain the machine text corresponding to the question text of each of the aforementioned popular dialogues; The response text of each popular dialogue and the machine text corresponding to each popular dialogue are input into the embedding layer to obtain the fourth embedding feature vector and the fifth embedding feature vector; Calculate the second similarity between the fourth embedded feature vector and the fifth embedded feature vector; Select target dialogues from the popular dialogues whose second similarity is less than the second similarity threshold.

6. An information processing device, characterized in that, include: The acquisition module is used to classify the intent of the original dialogue using a pre-trained classification model to obtain the intent category of each original dialogue, and to acquire multiple candidate dialogues from multiple original dialogues whose intent category is a valid question answer. The merging module is used to merge multiple candidate dialogues based on the correlation between the question texts of adjacent candidate dialogues to obtain multiple merged dialogues. Specifically, the question texts of the first two adjacent candidate dialogues are input into an embedding layer to obtain a first embedded feature vector and a second embedded feature vector; a first similarity between the first embedded feature vector and the second embedded feature vector is calculated; if the first similarity is greater than or equal to a first similarity threshold, any two adjacent candidate dialogues are merged; and the question texts of subsequent adjacent candidate dialogues are merged sequentially to obtain multiple merged dialogues. The clustering module is used to cluster multiple merged dialogues to obtain at least one group of dialogue clusters; The first selection module is used to calculate the popularity value of each dialogue in the target cluster dialogue group based on the frequency of similar questions, the negative feedback rate of similar questions, and the cluster center distance for each dialogue in the target cluster dialogue group, and select the dialogue with the highest popularity value from the target cluster dialogue group as a popular dialogue; wherein, the target cluster dialogue group can be any dialogue group in the cluster dialogue group; The second selection module selects target dialogues that meet the similarity criteria from the popular dialogues based on the similarity between the response text of each popular dialogue and the machine text corresponding to each popular dialogue, and stores the target dialogues in the dialogue knowledge base.

7. The apparatus according to claim 6, characterized in that, The step of classifying the intent of the original dialogue using a pre-trained classification model to obtain the intent category for each original dialogue includes: Input the question text, entity information, syntactic features, and question text of historical dialogues for each original dialogue into the trained classification model to obtain the intent category of each original dialogue.

8. The apparatus according to claim 7, characterized in that, The classification model includes an embedding layer, a deep neural network layer, and an activation function layer. The acquisition module, when inputting entity information, syntactic features, and historical dialogues of each original dialogue into the trained classification model to obtain the intent category of each original dialogue, is used for: The question text of each original dialogue and the question text of each historical dialogue of the original dialogue are input into the embedding layer to obtain the first dialogue feature; The entity information and syntactic features of each original dialogue are input into a deep neural network layer to obtain the second dialogue features; The first dialogue feature and the second dialogue feature are input into the activation function layer to obtain the intent category of each original dialogue.

9. The apparatus according to claim 6, characterized in that, When the clustering module clusters multiple merged dialogues to obtain at least one group of dialogue clusters, it is used to: The question text and response text of each merged dialogue are concatenated to obtain the concatenated text of each merged dialogue; The concatenated text of each merged dialogue is input into the embedding layer to obtain the third embedding feature vector of each merged dialogue; Cluster the third embedding feature vectors of multiple merged dialogues to obtain the cluster to which each merged dialogue belongs; Grouping is performed based on the cluster to which each merged dialogue belongs, resulting in at least one clustered dialogue group.

10. The apparatus according to claim 6, characterized in that, When the second selection module selects a target dialogue that meets the similarity condition from the popular dialogues based on the similarity between the response text of each popular dialogue and the machine text corresponding to each popular dialogue, it is used to: Obtain the machine text corresponding to the question text of each of the aforementioned popular dialogues; The response text of each popular dialogue and the machine text corresponding to each popular dialogue are input into the embedding layer to obtain the fourth embedding feature vector and the fifth embedding feature vector; Calculate the second similarity between the fourth embedded feature vector and the fifth embedded feature vector; Select target dialogues from the popular dialogues whose second similarity is less than the second similarity threshold.

11. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the steps of the method as described in any one of claims 1-5.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-5.

13. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-5.

Images