Large model-based conversational knowledge management intelligent agent assistance method and device and electronic equipment

Abstract

The application discloses a large model-based conversation knowledge management intelligent agent auxiliary method and device and electronic equipment. Embodiments of the application calculate the first correlation between each word in the group message to determine at least one theme word in the group message, while calculating the second correlation between each conversation message and at least one conversation message adjacent in time sequence to determine the associated message of each conversation message, thereby forming a theme conversation group for a specific theme, calculating the third correlation of each word vector of the plurality of conversation messages contained therein to determine the first related word of each theme word, calculating the fourth correlation of the first related word and the word adjacent in position to determine the second related word associated therewith, and finally generating a theme sequence for the corresponding theme word, which can greatly improve the efficiency and accuracy of group message management, enabling group members to quickly obtain key information and know to-do items.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of text processing, and in particular to a large model-based conversation knowledge management agent assisting method and device and electronic equipment. BACKGROUND

[0002] In the current era of digital information explosion, social network services (SNS) and various online communication tools are widely used in personal and enterprise communication. Especially in social groups and work groups, the number of messages is huge and the content is complex. Group managers often need to spend a lot of time to filter out key information such as to-do lists, important notifications or decision-making basis. However, due to information overload, manual filtering is inefficient and easy to miss important content.

[0003] Existing message management technologies have limitations in handling such complex scenarios, and it is difficult to meet the needs of efficient and accurate knowledge management and information aggregation of group messages. For example, traditional keyword search methods cannot understand the context semantics, resulting in inaccurate extraction results. In addition, the information in the group message usually exists in an unstructured form, containing a large amount of fragmented content, redundant information and even noise, further increasing the complexity of information extraction.

[0004] Therefore, there is an urgent need for a large model-based intelligent management scheme for group messages, which can use natural language processing (NLP) technology and deep learning models to extract meaningful information from massive unstructured messages. SUMMARY

[0005] The embodiments of the present application provide a large model-based conversation knowledge management agent assisting method and device and electronic equipment to solve the defect that the existing technology only manages group messages based on keywords, which is low in efficiency.

[0006] To achieve the above technical purpose, the embodiments of the present application propose a large model-based conversation knowledge management agent assisting method, comprising:

[0007] Obtaining a first group message, wherein the first group message contains a plurality of conversation messages sent by a plurality of group members;

[0008] Performing word segmentation processing on each conversation message contained in the first group message to generate a word vector of each word contained in each conversation message;

[0009] Based on the word vector and the position information of each word in the corresponding conversation message and the time sequence of the corresponding conversation message in the first group message, calculating the first correlation between the words;

[0010] According to the first correlation, determining at least one theme word in the first group message;

[0011] calculating a second correlation between each session message contained in the first group message and at least one session message adjacent to the session message in time sequence;

[0012] determining, according to the second correlation, associated messages of each session message containing the topic word to form at least one topic session group, wherein each topic session group corresponds to a topic word, and the topic session group contains a plurality of session messages continuous or discontinuous in time sequence;

[0013] calculating, for each topic session group in the at least one topic session group, a third correlation of word vectors of a plurality of session messages contained in the topic session group, wherein the third correlation represents semantic correlation between a word vector in a session message and each word vector in other session messages in the topic session group;

[0014] determining, according to the third correlation, a first related word corresponding to each topic word;

[0015] calculating a fourth correlation of the first related word with a word adjacent to the first related word in the respective session message;

[0016] determining, according to the fourth correlation, a second related word associated with the first related word;

[0017] generating a topic sequence for the corresponding topic word based on the first related word and the second related word.

[0018] According to an embodiment of the present application, the session message includes sender information, receiver information, sending time information and sending content.

[0019] According to an embodiment of the present application, obtaining the first group message includes: determining, for each session message in the obtained first group message, a data type to which the session message belongs; and converting the session message into text data according to the data type using a corresponding machine learning model.

[0020] According to an embodiment of the present application, converting the session message into text data according to the data type using a corresponding machine learning model includes: when it is determined that the session message belongs to an image data type, extracting at least one image feature of the session message using an image recognition model; determining at least one object contained in the session message according to the image feature; determining a text description corresponding to the session message according to object information of the object; and using the text description as the text data.

[0021] According to an embodiment of the present application, converting the data into text data according to the data type using a corresponding machine learning model comprises: when it is determined that the conversation message belongs to a voice data type, using a speech recognition model to convert the conversation message into literal text; using the literal text as the text data.

[0022] According to an embodiment of the present application, the large model-based conversation knowledge management intelligent agent assisting method further comprises:

[0023] The word vector calculation for each word in each conversation message in the topic conversation group includes a first result containing the position feature of the topic word corresponding to the topic conversation group in the conversation message and a second result containing the context feature of the topic word in the topic conversation group;

[0024] Feature fusion calculation is performed on the first result and the second result to generate fusion features;

[0025] According to the fusion features, the view tendency of the conversation message for the topic word is calculated; and

[0026] According to the view tendency, the topic sequence corresponding to the topic conversation group is updated.

[0027] An embodiment of the present application also provides a large model-based conversation knowledge management intelligent agent assisting device, comprising:

[0028] An acquisition module is configured to acquire a first group message, wherein the first group message contains a plurality of conversation messages sent by a plurality of group members;

[0029] A preprocessing module is configured to perform word segmentation processing on each conversation message contained in the first group message to generate a word vector of each word contained in each conversation message;

[0030] A first calculation module is configured to calculate a first correlation between words based on the word vector and the position information of each corresponding word in the corresponding conversation message and the time sequence of the corresponding conversation message in the first group message;

[0031] A first determination module is configured to determine at least one topic word in the first group message according to the first correlation;

[0032] A second calculation module is configured to calculate a second correlation between each conversation message contained in the first group message and at least one conversation message adjacent to it in the time sequence;

[0033] a second determining module configured to determine, according to the second correlation, associated messages of each session message containing the topic word to form at least one topic session group, wherein each topic session group corresponds to a topic word, and the topic session group contains a plurality of session messages which are continuous or discontinuous in time sequence;

[0034] a third calculating module configured to calculate, for each topic session group in the at least one topic session group, a third correlation of word vectors of the plurality of session messages contained therein, wherein the third correlation represents semantic correlation between a word vector in a session message and each word vector in other session messages in the topic session group;

[0035] a third determining module configured to determine, according to the third correlation, a first related word corresponding to each topic word;

[0036] a fourth calculating module configured to calculate a fourth correlation of the first related word with a word adjacent in position in each session message;

[0037] a fourth determining module configured to determine, according to the fourth correlation, a second related word associated with the first related word;

[0038] a generating module configured to generate a topic sequence for the corresponding topic word based on the first related word and the second related word.

[0039] Embodiments of the present application further provide an electronic device, comprising:

[0040] a memory configured to store a program;

[0041] a processor configured to execute the program stored in the memory to perform the method for assisting a large model-based session knowledge management agent according to the embodiments of the present application.

[0042] Embodiments of the present application further provide a computer readable storage medium having stored thereon a computer program executable by a processor, wherein the program, when executed by the processor, implements the method for assisting a large model-based session knowledge management agent as provided by the embodiments of the present application.

[0043] According to the method and device for assisting a large model-based conversation knowledge management agent and the electronic device, the word vectors of each word contained in each conversation message in the group message are obtained by performing word segmentation on each conversation message contained in the group message, the first correlation between the words is calculated to determine at least one theme word in the group message, the second correlation between each conversation message and at least one conversation message adjacent in time sequence is calculated to determine the associated messages of each conversation message, thereby forming a theme conversation group for a specific theme, the third correlation of the word vectors of the plurality of conversation messages contained in each theme conversation group is calculated to determine the first related words of each theme word, and the fourth correlation of the first related words and the words adjacent in position is further calculated to determine the second related words associated therewith, and finally, the theme sequence for the corresponding theme word can be generated based on the first related words and the second related words. Therefore, the theme words related to a plurality of words can be determined from the group message containing a large number of conversation messages, each conversation message is re-clustered to form a theme conversation group, and then the first related words closely related to the theme words and the second related words associated with the first related words can be determined by calculating the third correlation and the fourth correlation in the theme conversation group, so that the expressions closely related to the theme words in the conversation message can be effectively extracted to form a theme sequence, greatly improving the efficiency and accuracy of group message management, and enabling group members to quickly obtain key information and know to-do items.

[0044] The above description is only a summary of the technical solutions of the present application. In order to enable the technical means of the present application to be more clearly understood, and to be implemented according to the content of the description, and in order to enable the above and other purposes, features and advantages of the present application to be more apparent and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0045] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not meant to limit the present application. Moreover, the same reference numerals in the accompanying drawings indicate the same or similar components. In the drawings:

[0046] Figure 1 a flowchart of an embodiment of the large model-based conversation knowledge management agent assistance method provided by the present application;

[0047] Figure 2 a structural schematic diagram of an embodiment of the large model-based conversation knowledge management agent assistance device provided by the present application;

[0048] Figure 3 a structural schematic diagram of an embodiment of the electronic device provided by the present application. DETAILED DESCRIPTION

[0049] Exemplary embodiments of the present application will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

[0050] Embodiment One

[0051] In the current era of information explosion, social network services (SNS) and online communication tools are widely used in personal and business communication. However, the number of messages in social groups and work groups is huge, and the content is complex, and managers need to spend a lot of time screening out key information, such as to-do lists, important notifications or decision-making basis. Due to information overload, manual screening is inefficient and prone to missing important content.

[0052] Existing message management techniques have limitations in handling complex scenarios. Traditional keyword search methods cannot understand the context semantics, resulting in inaccurate extraction results. In addition, group messages are usually in unstructured form, containing a large amount of fragmented content, redundant information and noise, which further increases the difficulty of information extraction.

[0053] Therefore, there is an urgent need for an intelligent large model-based conversation knowledge management agent assisted solution based on a large model. Through natural language processing (NLP) technology and deep learning models, this solution can extract meaningful information from a large amount of unstructured messages. It can quickly and accurately filter out key content, structure and summarize information, thereby improving management efficiency, reducing information omission, and meeting the needs of modern information management.

[0054] As shown in Figure 1 , the method for assisting a large model-based conversation knowledge management agent according to an embodiment of the present application can include: Figure 1 is a flowchart illustrating one embodiment of a method for assisting a large model-based conversation knowledge management agent according to an embodiment of the present application. The method for assisting a large model-based conversation knowledge management agent according to an embodiment of the present application can include:

[0055] S101, obtaining a first group message.

[0056] In an embodiment of the present application, the first group message can be obtained in step S101, which can contain a plurality of conversation messages sent by a plurality of group members. For example, various messages in the corresponding instant communication tool can be obtained using the API interface provided by various instant communication platforms, and the group message of the specified group can be further obtained as the first group message.

[0057] The session message included in the first group message can include sender information, receiver information, sending time information, sending content, and the like, and in the embodiments of the present application, the sending content of the session message can have various data formats, such as text data, image data, or voice data. Therefore, in the embodiments of the present application, the data type to which each session message belongs can also be determined for each session message in the first group message obtained in step S101. For example, a pre-trained machine learning model can be used to calculate the relevance between each session message, such as the sending content included in the session message, and a preset data type, and thereby determine the data type to which the sending content included in the drawing message belongs based on the relevance, and then the corresponding machine learning model can be used to convert it into text data according to the data type thus determined.

[0058] For example, when it is determined that the session message (in which the sending content is included) belongs to the image data type, at least one image feature of the session message can be extracted using an image recognition model in step S101, and then at least one object included in the session message can be determined according to the image features. For example, it can be determined from the image features that the session message or the sending content included therein contains a figure of a person and a figure of a car. That is, in the embodiments of the present application, a user can send a picture of a person and a car that he or she takes as sending content in a group, so that when it is determined that this session message is of the image data type in step S101, image recognition model can be used to extract various image features therefrom, and further determine that a person and a car are included therein. In the embodiments of the present application, when it is determined that a person is included in the image, a face recognition model can be further used to determine the identity of the person, or when it is determined that a vehicle is included in the image, a vehicle recognition model can be further used to determine the model of the vehicle, and the like, without being limited thereto.

[0059] After the object included in the sending content as image data is determined, the text description corresponding to the session message can be further determined according to the object information of the object, and the text description is used as the text data of the session message.

[0060] In addition, when it is determined that the session message (in which the sending content is included) belongs to the voice data type, a voice recognition model can be used to convert the session message into a literal text in step S101, and then the literal text is used as the text data of the session message.

[0061] S102, for each session message included in the first group message, performing word segmentation processing to generate a word vector of each word included in each session message.

[0062] In step S102, word segmentation processing can be performed on each conversation message in the first group message obtained in step S101. For example, the conversation message of the text data obtained in step S101 can be taken as the first group message, and word segmentation processing can be directly performed on the text data in step S102 to generate a word vector of each word contained in the conversation message.

[0063] In the embodiments of the present application, word embedding processing (WE) can be used to convert each word contained in the conversation message into a word vector, so that further processing can be performed in subsequent steps. In particular, in the embodiments of the present application, sentiment-specific word embedding processing (SSWE) can also be used in step S102 to convert the text data obtained in step S101 into a word vector, so that a word vector with sentiment information can be obtained, which can better reflect the sentiment of the words in the text data.

[0064] In step S103, based on the word vector and the position information of each word in the corresponding conversation message and the time sequence of the corresponding conversation message in the first group message, a first correlation between the words is calculated.

[0065] In step S104, at least one theme word in the first group message is determined according to the first correlation.

[0066] In step S103, the first correlation between the words can be calculated for the word vectors of the words obtained after the word segmentation processing of each conversation message in step S102. Specifically, the first correlation between the words can be calculated using the word vector of each word, the position information of the word in the conversation message, and the time sequence of the conversation message in the first group message. For example, the position information and the time sequence can be used as the weight values of the corresponding words to calculate the first correlation between the words. Then the theme word in the first group message can be determined according to the first correlation. In the embodiments of the present application, the correlation of each word obtained in step S103 can be sorted according to the value, and one or more words with the highest correlation are determined as the theme word. That is, these words have relevance with the words in the multiple conversation messages, and therefore, it is likely that the members in the group were discussing the theme when the group message was obtained in step S101.

[0067] In step S105, a second correlation between each conversation message contained in the first group message and at least one conversation message adjacent to it in time sequence is calculated.

[0068] S106, according to the second correlation, determine the associated messages of the session messages containing the topic word to form at least one topic session group.

[0069] After step S104, the second correlation between each session message contained in the first group message and the session message adjacent in time can be calculated in step S105. In step S105, the correlation between two session messages can be calculated as the second correlation in the whole session message (i.e. the sending content contained therein). Then in step S106, the associated messages of the session messages containing the topic word determined in step S104, i.e. the other session messages having the correlation with the session messages containing the topic word, can be determined based on the second correlation determined in step S105. In the embodiments of the present application, when the group members discuss a certain topic, they can also discuss other things in the process of discussion, therefore, in step S106, the correlation calculated in step S105 can be used to select the messages related to the session messages containing the topic word, i.e. the related session messages are likely to represent that the group users are discussing the topic, instead of having started to discuss other topics. Each topic session group determined in step S106 can correspond to a topic word, and can also correspond to multiple topic words. Therefore, in the topic session group determined in step S106, multiple session messages continuous or discontinuous in time sequence can be contained. That is, among the session messages contained in the group messages obtained in step S101, it is likely that two or more group members start to discuss other things, even chat, therefore, the correlation between the session messages calculated in step S105 can be used to remove these session messages irrelevant to the current topic, so as to form the topic session group containing only the topic word or the session messages containing the topic word.

[0070] S107, for each topic session group in the at least one topic session group, calculate the third correlation of each word vector of the multiple session messages contained therein.

[0071] S108, according to the third correlation, determine the first related word corresponding to each topic word.

[0072] In step S107, the third correlation between the word vectors of the words in the plurality of conversation messages contained in the topic conversation group determined in step S106 can be calculated. In the embodiments of the present application, since the conversation messages irrelevant to the topic have been removed from the group messages by the processing in step S106, the conversation messages with higher relevance to the topic thus aggregated in step S107 can be recalculated, and further the correlation between the words contained therein can be calculated. In the embodiments of the present application, the third correlation can represent the semantic correlation between the word vector in a conversation message and the word vectors in other conversation messages in the topic conversation group. In step S108, the first relevant words corresponding to the topic words can be determined based on such correlation. For example, the words with relevance to the topic words higher than a preset threshold can be calculated as the first relevant words.

[0073] In step S109, the fourth correlation between the first relevant words and the words adjacent in position in the respective conversation messages can be calculated.

[0074] In step S110, the second relevant words associated with the first relevant words can be determined according to the fourth correlation.

[0075] In step S111, the topic sequence for the corresponding topic words can be generated based on the first relevant words and the second relevant words.

[0076] In step S109, the other relevant words related to the relevant words determined in step S108 can be further calculated. For example, when the topic discussed in the group is project A, one or more relevant words related to the topic word project, such as progress, problem, etc., can be calculated as the first relevant words in step S108, and then in step S109, other words adjacent in position to the first relevant words can be further determined according to the first relevant words determined in step S108, since in the conversation process, the participants are likely to continue to express the specific circumstances of the project progress after mentioning the project progress, and thus such specific description words can be determined by calculating the correlation between the words adjacent in position to the first relevant words and the first relevant words, so that in step S111, the topic sequence for the corresponding topic can be generated based on the first relevant words and the second relevant words thus determined, to be displayed to the relevant user, which helps the user to have a clearer understanding of the conversation content in the group messages related to the specified topic.

[0077] In addition, in the embodiment of the present application, the large model based conversation knowledge management intelligent agent assisting method can further calculate the emotion or attitude represented by each conversation message in the topic conversation group. For example, the first result containing the position feature of the topic word corresponding to the topic conversation group in the conversation message and the second result containing the context feature of the topic word in the topic conversation group can be calculated for the word vector of each word in each conversation message in the topic conversation group; the first result and the second result are calculated by feature fusion to generate fusion features; the view tendency of the conversation message to the topic word is calculated according to the fusion features; and the topic sequence generated in step S111 corresponding to the topic conversation group is updated according to the view tendency. For example, by the above-mentioned manner, some discussion contents for the topic can be screened out, and different annotations or display effects can be given to the related words in the topic sequence based on the calculated view tendency, so as to make the user more clearly understand the attitude of the conversation discussion related to the topic.

[0078] Therefore, according to the large model based conversation knowledge management intelligent agent assisting method, by performing word segmentation on each conversation message contained in the group message to obtain the word vector of each word contained therein, calculating the first correlation between the words to determine at least one topic word in the group message, and calculating the second correlation between each conversation message and at least one conversation message adjacent in time sequence to determine the associated message of each conversation message, a topic conversation group for a specific topic is formed, the third correlation of the word vectors of the multiple conversation messages contained in each topic conversation group is calculated to determine the first related word of each topic word, and the fourth correlation between the first related word and the word adjacent in position is further calculated to determine the second related word associated therewith, and finally the topic sequence for the corresponding topic word can be generated based on the first related word and the second related word. Therefore, the topic word related to multiple words can be determined from the group message containing a large number of conversation messages, so that each conversation message is re-clustered to form each topic conversation group, and then the first related word closely related to the topic word and the second related word associated with the first related word can be determined in the topic conversation group by calculating the third correlation and the fourth correlation, so that the expression closely related to the topic word in the conversation message can be effectively extracted to form the topic sequence, greatly improving the efficiency and accuracy of group message management, and enabling the group members to quickly obtain key information and know the to-do list.

[0079] Embodiment Two

[0080] Figure 2 An embodiment of the large model based conversation knowledge management intelligent agent assisting device provided by the present application is shown in a structural schematic diagram. As shown in Figure 2As shown in the figure, the embodiment of the present application provides a large model-based conversation knowledge management intelligent agent auxiliary device, which can include an acquisition module 201, a preprocessing module 202, a first calculation module 203, a first determination module 204, a second calculation module 205, a second determination module 206, a third calculation module 207, a third determination module 208, a fourth calculation module 209, a fourth determination module 210, and a generation module 211.

[0081] The acquisition module 201 can be used to acquire the first group message.

[0082] In the embodiment of the present application, the acquisition module 201 can acquire the first group message, which can contain a plurality of conversation messages sent by a plurality of group members. For example, the API interface provided by various instant messaging platforms can be used to acquire various messages in the corresponding instant messaging tool, and the group message of the specified group can be further acquired as the first group message.

[0083] The conversation message contained in the first group message can include sender information, receiver information, sending time information, sending content, etc., and in the embodiment of the present application, the sending content of the conversation message can have various data formats, such as text data, image data or voice data. Therefore, in the embodiment of the present application, the acquisition module 201 can further determine the data type of each conversation message in the acquired first group message. For example, a pre-trained machine learning model can be used to calculate the association between each conversation message, such as the sending content contained in the conversation message, and the preset data type, and thereby determine the data type of the sending content contained in the drawing message based on the association, and then the corresponding machine learning model can be used to convert it to text data according to the data type thus determined.

[0084] For example, when it is determined that the session message (the sending content contained therein) belongs to the image data type, the obtaining module 201 can use an image recognition model to extract at least one image feature of the session message, and then determine at least one object contained in the session message according to the image features. For example, it can be determined according to the image features that the session message or the sending content contained therein contains a figure of a person and a figure of a car. That is, in the embodiment of the present application, the user can send a picture of a person and a car taken by himself as the sending content in the group, so that when it is determined that the session message is of the image data type, the obtaining module 201 can use the image recognition model to extract various image features therefrom, and further determine that it contains a person and a car. In the embodiment of the present application, when it is determined that the image contains a person, a face recognition model can be further used to determine the identity of the person, or when it is determined that the image contains a vehicle, a vehicle recognition model can be further used to determine the model of the vehicle, and the like, which are not limited in the present application.

[0085] After determining the object contained in the sending content as image data, the text description corresponding to the session message can be further determined according to the object information of the object, and the text description is used as the text data of the session message.

[0086] In addition, when it is determined that the session message (the sending content contained therein) belongs to the voice data type, the obtaining module 201 can use a speech recognition model to convert the session message into a literal text, and then use the literal text as the text data of the session message.

[0087] The preprocessing module 202 is configured to perform word segmentation processing on each session message contained in the first group message to generate a word vector of each word contained in each session message.

[0088] The preprocessing module 202 can perform word segmentation processing on each session message in the first group message obtained by the obtaining module 201. For example, the session message of the text data that can be obtained in the obtaining module 201 is taken as the first group message, and the preprocessing module 202 can directly perform word segmentation processing on such text data to generate a word vector of the word contained in the session message.

[0089] In the embodiments of the present application, the word embedding processing (WE) can be used to convert each word contained in the conversation messages into a word vector, so that further processing can be performed in subsequent steps. In particular, in the embodiments of the present application, the preprocessing module 202 can further use the sentiment-specific word embedding processing (SSWE) to convert the text data obtained by the obtaining module 201 into a word vector, so as to obtain a word vector with sentiment information, which can better reflect the sentiment tendency of the words in the text data.

[0090] The first calculation module 203 is configured to calculate the first correlation between each word based on the word vector and the position information of each word in the corresponding conversation message and the time sequence of the corresponding conversation message in the first group message.

[0091] The first determination module 204 is configured to determine at least one theme word in the first group message according to the first correlation.

[0092] The first calculation module 203 can calculate the first correlation between each word based on the word vector of the word obtained by the preprocessing module 202 after performing the word segmentation processing on each conversation message. Specifically, the first correlation between each word can be calculated using the word vector of each word, the position information of the word in the conversation message, and the time sequence of the conversation message in the first group message. For example, the position information and the time sequence can be used as the weight value of the corresponding word to calculate the first correlation between the words. Then the theme word in the first group message can be determined according to the first correlation. In the embodiments of the present application, the correlation of each word obtained by the first calculation module 203 can be sorted according to the value, and one or more words with the highest correlation are determined as the theme word. That is, these words have relevance with the words in the plurality of conversation messages, and therefore, it is likely that when the obtaining module 201 obtains the group message, the members in the group are discussing the theme.

[0093] The second calculation module 205 is configured to calculate the second correlation between each conversation message contained in the first group message and at least one conversation message adjacent to it in time sequence.

[0094] The second determination module 206 is configured to determine the associated messages of each conversation message containing the theme word according to the second correlation, to form at least one theme conversation group.

[0095] The second calculating module 205 can be configured to calculate the second correlation between each conversation message included in the first group message and the conversation message adjacent in time. The second calculating module 205 can calculate the correlation between two conversation messages as the second correlation in the whole conversation message (i.e. the sending content included therein). Then the second determining module 206 can determine the associated message of the conversation message including the topic word determined by the first determining module 204, i.e. the other conversation message having the correlation with the conversation message including the topic word, based on the second correlation determined by the second calculating module 205. In the embodiments of the present application, when the group members discuss a certain topic, they can also discuss other things in the process of discussion, therefore the second determining module 206 can select the message related to the conversation message including the topic word by using the correlation calculated by the second calculating module 205, i.e. the conversation message related can probably represent that the group users are discussing the topic, instead of having started to discuss other topic. Each topic conversation group determined by the second determining module 206 can correspond to one topic word, and can also correspond to multiple topic words. Therefore, the topic conversation group determined by the second determining module 206 can include multiple conversation messages continuous or discontinuous in time sequence. That is, among the conversation messages included in the group message acquired by the acquiring module 201, there can be two or more group members starting to discuss other things, even chatting, therefore the correlation between the conversation messages can be calculated by the second calculating module 205 to remove the conversation messages irrelevant to the current topic, so as to form the topic conversation group including only the topic word or the conversation message including the topic word.

[0096] The third calculating module 207 is configured to calculate the third correlation of each word vector included in the multiple conversation messages in each topic conversation group in the at least one topic conversation group.

[0097] The third determining module 208 is configured to determine the first related word corresponding to each topic word according to the third correlation.

[0098] The third calculating module 207 can calculate a third correlation between word vectors of words in the plurality of conversation messages contained in the topic conversation group determined by the second determining module 206. In the embodiments of the present application, since the conversation messages irrelevant to the topic have been removed by the processing of the second determining module 206, the third calculating module 207 can recalculate the correlation between the words in the conversation messages with high relevance to the topic thus aggregated, and further calculate the correlation between the word vectors in the conversation messages. In the embodiments of the present application, the third correlation can represent the semantic correlation between the word vectors in a conversation message and the word vectors in other conversation messages in the topic conversation group. The third determining module 208 can determine the first relevant word corresponding to the topic word based on the correlation. For example, the words with a relevance higher than a preset threshold to the topic word can be determined as the first relevant word.

[0099] The fourth calculating module 209 is configured to calculate a fourth correlation between the first relevant word and a word adjacent to the first relevant word in the respective conversation message.

[0100] The fourth determining module 210 is configured to determine a second relevant word associated with the first relevant word according to the fourth correlation.

[0101] The generating module 211 is configured to generate a topic sequence for the corresponding topic word based on the first relevant word and the second relevant word.

[0102] The fourth calculating module 209 can further calculate other relevant words associated with the relevant word determined by the third determining module 208. For example, when the topic discussed in the group is project A, one or more relevant words associated with the topic word project, such as progress, problem, etc., can be determined as the first relevant word by the third determining module 208, and then the fourth calculating module 209 can further determine the correlation between other words adjacent to the first relevant word in the position sequence and the first relevant word according to the first relevant word determined by the third determining module 208, because the participants are likely to continue to express the specific situation of the project progress after mentioning the project progress in the conversation process, and thus the fourth calculating module 209 can determine such specific description words by calculating the correlation between the words adjacent to the first relevant word in the position sequence and the first relevant word. Thus, the generating module 211 can generate a topic sequence for the corresponding topic based on the first relevant word and the second relevant word thus determined, and display the topic sequence to the relevant user, which helps the user to have a clearer understanding of the conversation content related to the specified topic in the group message.

[0103] In addition, in the embodiment of the present application, the large model based conversation knowledge management intelligent agent auxiliary device can further calculate the emotion or attitude represented by each conversation message in the topic conversation group. For example, the first result containing the position feature of the topic word corresponding to the topic conversation group in the conversation message and the second result containing the context feature of the topic word in the topic conversation group can be calculated for the word vector of each word in each conversation message in the topic conversation group; the first result and the second result are calculated by feature fusion to generate fusion features; the view tendency of the conversation message to the topic word is calculated according to the fusion features; and the topic sequence generated in step S111 corresponding to the topic conversation group is updated according to the view tendency. For example, by the above-mentioned manner, some discussion contents for the topic can be screened out, and different annotations or display effects can be given to the related words in the topic sequence based on the view tendency calculated in this way, so that the user can more clearly understand the attitude of the conversation discussion related to the topic.

[0104] Therefore, according to the large model based conversation knowledge management intelligent agent auxiliary device of the embodiment of the present application, by performing word segmentation on each conversation message contained in the group message to obtain the word vector of each word contained therein, calculating the first correlation between the words to determine at least one topic word in the group message, and calculating the second correlation between each conversation message and at least one conversation message adjacent in time sequence to determine the associated message of each conversation message, a topic conversation group for a specific topic is formed, the third correlation of the word vectors of the multiple conversation messages contained in each topic conversation group is calculated to determine the first related word of each topic word, and the fourth correlation of the first related word and the word adjacent in position is further calculated to determine the second related word associated therewith, and finally the topic sequence for the corresponding topic word can be generated based on the first related word and the second related word. Therefore, the topic word related to multiple words can be determined from the group message containing a large number of conversation messages, so that each conversation message is re-clustered to form each topic conversation group, and then the first related word closely related to the topic word and the second related word associated with the first related word can be determined in the topic conversation group by calculating the third correlation and the fourth correlation, so that the expression closely related to the topic word in the conversation message can be effectively extracted to form the topic sequence, greatly improving the efficiency and accuracy of group message management, enabling group members to quickly obtain key information and know the to-do list.

[0105] Embodiment three

[0106] The internal functions and structures of the large model based conversation knowledge management intelligent agent auxiliary device are described above, and the device can be implemented as an electronic device. Figure 3 The structural schematic diagram of the electronic device embodiment provided in the present application is shown in the following figure. Figure 3As shown, the electronic device includes a memory 31 and a processor 32.

[0107] The memory 31 is used to store programs. In addition to the above programs, the memory 31 can also be configured to store various other data to support operations on the electronic device. Examples of these data include instructions for any application or method operating on the electronic device, contact data, phonebook data, messages, pictures, videos, etc.

[0108] The memory 31 can be implemented by any type of volatile or non-volatile storage devices 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.

[0109] The processor 32 is not limited to a central processing unit (CPU) but can also be a graphics processing unit (GPU), a field-programmable gate array (FPGA), an embedded neural processing unit (NPU), or an artificial intelligence (AI) chip, etc. The processor 32 is coupled to the memory 31 and executes programs stored in the memory 31. When the programs are run, the method for assisting the large model-based conversation knowledge management agent in the above embodiment I is executed.

[0110] Further, as shown, Figure 3 The electronic device can further include a communication component 33, a power supply component 34, an audio component 35, a display 36, and other components. Figure 3 Some components are only schematically shown in the electronic device, which does not mean that the electronic device only includes Figure 3 the components shown.

[0111] The communication component 33 is configured to facilitate wired or wireless communication between the electronic device and other devices. The electronic device can access a wireless network based on a communication standard, such as WiFi, 3G, 4G, or 5G, or a combination thereof. In an example embodiment, the communication component 33 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In an example embodiment, the communication component 33 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0112] The power supply component 34 provides power to various components of the electronic device. The power supply component 34 can include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device.

[0113] The audio component 35 is configured to output and / or input audio signals. For example, the audio component 35 includes a microphone (MIC) that is configured to receive an external audio signal when the electronic device is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal can be further stored in the memory 31 or transmitted via the communication component 33. In some embodiments, the audio component 35 also includes a speaker for outputting audio signals.

[0114] The display 36 includes a screen, which can 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 touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense a touch, a slide, and a gesture on the touch panel. The touch sensor can not only detect a boundary of a touching or a sliding action, but also detect duration and pressure related to the touching or sliding action.

[0115] It is understood by those skilled in the art that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program, when executed, performs steps including the above-mentioned method embodiments; and the foregoing storage medium includes ROM, RAM, magnetic disk or optical disk and various storage media that can store program codes.

[0116] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for assisting intelligent agents in conversational knowledge management based on a large model, characterized in that, include: Obtain a first group message, wherein the first group message contains multiple session messages sent by multiple group members; Each session message contained in the first group message is segmented into words to generate word vectors for each word contained in each session message; Based on the word vectors, the position information of each word in the corresponding conversation message, and the time order of the corresponding conversation message in the first group message, the first correlation between each word is calculated. Based on the first correlation, at least one topic word in the first group message is determined; Calculate a second correlation between each session message contained in the first group message and at least one session message that is adjacent to it in time sequence; Based on the second correlation, the associated messages of each session message containing the topic word are determined to form at least one topic session group, wherein each topic session group corresponds to a topic word, and the topic session group contains multiple session messages that are consecutive or non-consecutive in time order. For each topic session group in the at least one topic session group, calculate the third relevance of each word vector of the multiple session messages contained therein, wherein the third relevance represents the semantic relevance between the word vector in a session message and the word vectors in other session messages in the topic session group; Based on the third relevance, determine the first related words corresponding to each topic word; Calculate the fourth relevance of the first related word to the words that are adjacent in position in their respective conversation messages; Based on the fourth correlation, determine the second related word associated with the first related word; Generate a topic sequence for the corresponding topic words based on the first related words and the second related words.

2. The method for assisting intelligent agents in conversational knowledge management based on a large model according to claim 1, characterized in that, The session message includes sender information, receiver information, sending time information, and sending content.

3. The method for assisting intelligent agents in conversational knowledge management based on a large model according to claim 1, characterized in that, The process of obtaining the first group message includes: For each session message in the first group of messages obtained, determine the data type of each session message; Based on the data type, the corresponding machine learning model is used to convert it into text data.

4. The method for assisting intelligent agents in conversational knowledge management based on a large model according to claim 3, characterized in that, The step of converting the data type into text data using the corresponding machine learning model includes: When it is determined that the session message belongs to the image data type, at least one image feature of the session message is extracted using an image recognition model; Based on the image features, at least one object contained in the session message is determined; Determine the text description corresponding to the session message based on the object information of the object; The text description is used as the text data.

5. The method for assisting intelligent agents in conversational knowledge management based on a large model according to claim 3, characterized in that, The step of converting the data type into text data using the corresponding machine learning model includes: When it is determined that the conversation message belongs to the voice data type, the speech recognition model is used to convert the conversation message into text. The text is used as the text data.

6. The method for assisting intelligent agents in conversational knowledge management based on a large model according to claim 1, characterized in that, The method further includes: For each word in each session message of the topic session group, the word vector calculation includes a first result containing the positional features of the topic word corresponding to the topic session group in the session message and a second result containing the contextual features of the topic word in the topic session group; Feature fusion calculation is performed on the first result and the second result to generate fused features; Calculate the viewpoint bias of the conversation message regarding the topic words based on the fusion features; and The topic sequence corresponding to the topic session group is updated based on the stated viewpoint.

7. A conversational knowledge management intelligent agent auxiliary device based on a large model, characterized in that, include: The acquisition module is used to acquire a first group message, wherein the first group message contains multiple session messages sent by multiple group members; The preprocessing module is used to perform word segmentation on each session message contained in the first group message to generate word vectors for each word contained in each session message; The first calculation module is used to calculate the first correlation between words based on the word vectors, the position information of each word in the corresponding session message, and the time order of the corresponding session message in the first group message. The first determining module is configured to determine at least one topic word in the first group message based on the first correlation. The second calculation module is used to calculate a second correlation between each session message contained in the first group message and at least one session message that is adjacent to it in time sequence; The second determining module is used to determine the associated messages of each session message containing the topic word according to the second correlation, so as to form at least one topic session group, wherein each topic session group corresponds to a topic word, and the topic session group contains multiple session messages that are consecutive or non-consecutive in time order. The third calculation module is used to calculate the third relevance of each word vector of a plurality of conversation messages contained in each topic conversation group for the at least one topic conversation group, wherein the third relevance represents the semantic relevance between the word vector in a conversation message and the word vectors in other conversation messages in the topic conversation group. The third determining module is used to determine the first related words corresponding to each topic word based on the third relevance; The fourth calculation module is used to calculate the fourth relevance of the first related word to the words that are adjacent in position in their respective conversation messages; The fourth determining module is used to determine a second related word associated with the first related word based on the fourth relevance; The generation module is used to generate a topic sequence for the corresponding topic words based on the first related words and the second related words.

8. The large-model-based conversational knowledge management intelligent agent auxiliary device according to claim 7, characterized in that, The acquisition module is further used for: For each session message in the first group of messages obtained, determine the data type of each session message; Based on the data type, the corresponding machine learning model is used to convert it into text data.

9. An electronic device, characterized in that, include: Memory, used to store programs; A processor is configured to run the program stored in the memory to execute the large-model-based intelligent agent-assisted method for conversational knowledge management as described in any one of claims 1 to 6.

Images