Chat robot data processing method and device based on large model

Abstract

The invention provides a chat robot data processing method and device based on a large model, and relates to the technical field of large model application, and the method comprises the steps: calling a chat robot to carry out the matching of a real-time news event and a chat record of a chat group, and determining the historical chat record content associated with the real-time news event; judging whether the historical chat record content contains historical request information for the real-time news event or not; if not, inputting the historical chat record content and the current request information into a multi-mode large model, and outputting a response for the current request information; if so, fusing the historical information and the current information corresponding to the historical request information by using a multi-modal large model; inputting the fusion information and the historical chat record content into a multi-mode large model, and outputting a response for the current request information; through the fusion information of the historical questions for the real-time news events, the technical problem that the response accuracy of real-time news is not high in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the field of large model application technology, and in particular to a chat robot data processing method and device based on a large model. Background Art

[0002] With the emergence of multimodal large models (multimodal large language models), AI can communicate smoothly with people and has shown leaps and bounds in semantic understanding and logical reasoning.

[0003] The inventors have found that the large model can be applied to the chatbot of the group, so that group users can call the large model to ask questions during the conversation. However, the current multimodal large model generally obtains answers to user questions through a preset and updateable knowledge base; however, such answers are lagging and cannot meet the user's question requirements for real-time news events, that is, the accuracy of the answers to real-time news is not high. Summary of the invention

[0004] The purpose of the present invention is to provide a chat robot data processing method and device based on a large model, which can alleviate the technical problem of low accuracy in responding to real-time news in the prior art by fusing information of historical questions about real-time news events.

[0005] In a first aspect, an embodiment of the present invention provides a chatbot data processing method based on a large model, comprising: In response to current request information of a member in a chat group for a real-time news event, calling a chat robot; wherein the chat group includes at least one of the chat robots; The chat robot matches the real-time news event with the chat records of the chat group to determine the historical chat record content associated with the real-time news event; wherein the chat robot includes a multimodal large model; If the historical chat record content does not contain historical request information for the real-time news event, the historical chat record content and the current request information are input into the multimodal macro model, and a response to the current request information is output; wherein the historical request information is associated with the current request information; If the historical chat record content contains historical request information for real-time news events, the historical summary information and historical responses corresponding to the historical request information, as well as the current summary information and current responses corresponding to the historical request information are determined respectively, and input into the multimodal large model to output fusion information for the historical request information; The current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content are input into the multimodal big model, and a response to the current request information is output.

[0006] In combination with the first aspect, an embodiment of the present invention provides a first possible implementation of the first aspect, wherein the step of invoking a chat robot in response to a current request information of a member in a chat group for a real-time news event includes: In response to a preset target word input operation of a chat robot by a member of the chat group, triggering a working state of the chat robot; In response to current request information input by a member of the chat group, the chat robot is controlled to determine at least one corresponding real-time news event and summary information corresponding to each real-time news event according to the current request information.

[0007] In combination with the first aspect, an embodiment of the present invention provides a second possible implementation of the first aspect, wherein the step of matching the real-time news event with the chat record of the chat group by the chat robot to determine the historical chat record content associated with the real-time news event includes: Extracting keywords from the real-time news event by using the chat robot; Based on the extracted keywords, the chat records of the chat group are searched to determine whether there is historical chat record content that matches the real-time news event.

[0008] In combination with the first aspect, an embodiment of the present invention provides a third possible implementation of the first aspect, wherein the step of matching the real-time news event with the chat record of the chat group by the chat robot to determine the historical chat record content associated with the real-time news event includes: Classifying the chat records of the chat group according to the preset target words to obtain a first chat group for representing the chat between the group member and the chat robot and a second chat group for representing the chat between the group member and the group member; Extracting keywords from the real-time news event by using the chat robot; searching for matching chat record content from the first chat group and the second chat group respectively based on the extracted keywords, and determining a matching degree of each matching chat record content; Allocating corresponding weights of the first chat group and the second chat group according to the number of the matching chat record contents respectively existing in the first chat group and the second chat group; Determine the relevance of each of the matched chat record contents to the real-time news event based on the matching degree of each of the matched chat record contents and the corresponding weight of the chat group to which each of the matched chat record contents belongs; By comparing the correlation with a preset correlation threshold, it is determined whether there is historical chat record content that matches the real-time news event.

[0009] In combination with the first aspect, an embodiment of the present invention provides a fourth possible implementation of the first aspect, wherein the step of matching the real-time news event with the chat record of the chat group by the chat robot to determine the historical chat record content associated with the real-time news event includes: The chat robot performs semantic recognition on the chat records to determine the subject corresponding to each chat record content; Determining the relevance of each chat record content based on the matching degree of each chat record content corresponding to the topic; Extracting keywords from the real-time news event by using the chat robot; Searching the extracted keywords from each chat record of the chat group to determine a first chat record; The historical chat record content matching the real-time news event is jointly determined based on the second chat record and the first chat record whose relevance to the first chat record meets a preset relevance threshold.

[0010] In combination with the first aspect, an embodiment of the present invention provides a fifth possible implementation of the first aspect, wherein, if the historical chat record content does not contain historical request information for real-time news events, the step of inputting the historical chat record content and the current request information into a multimodal large model and outputting a response to the current request information includes: Determining whether the historical chat record content contains historical request information for real-time news events; If not, perform the following steps: Controlling the real-time search engine to retrieve at least one real-time news event corresponding to the current request information; Extracting corresponding key information from each of the real-time news events, and inputting the key information into the multimodal big model, and determining summary information corresponding to each of the real-time news events; wherein the multimodal big model includes a first preset template for extracting the summary information, and the first preset template is determined according to the format requirements and attribute requirements of the summary information; Determining the authenticity of the real-time news event corresponding to each of the summary information according to the verification rules and verification weights corresponding to each news element of the summary information in the third preset template; Determining a target real-time news event corresponding to the current request information from the real-time news events according to the authenticity of the real-time news event and the summary information; The current request information, the historical chat record content and the summary information of the target real-time news event are input into the multimodal large model, and a response to the current request information is output under the action of a second preset template.

[0011] In combination with the first aspect, an embodiment of the present invention provides a sixth possible implementation of the first aspect, wherein if the historical chat record content includes historical request information for a real-time news event, the steps of respectively determining historical summary information and historical responses corresponding to the historical request information, and current summary information and current responses corresponding to the historical request information, and inputting them into a multimodal large model to output fusion information for the historical request information include: Determining whether the historical chat record content contains historical request information for real-time news events; If included, perform the following steps: Acquiring, by the chat robot, historical summary information and historical responses corresponding to the historical request information from the historical chat record content; The historical summary information, the historical response, and the current summary of at least one latest real-time news event corresponding to the historical request information at the current moment are input into a multimodal large model, and fusion information for the historical request information is output.

[0012] In combination with the first aspect, an embodiment of the present invention provides a seventh possible implementation of the first aspect, wherein the steps of inputting the historical summary information, the historical response, and the current summary of at least one latest real-time news event corresponding to the historical request information at the current moment into the multimodal big model, and outputting fusion information for the historical request information include: Control the real-time search engine to retrieve and extract at least one latest real-time news event corresponding to the historical request information at the current moment; Extracting corresponding key information from each of the latest real-time news events, and inputting the key information into the multimodal big model, and determining the current summary information corresponding to each of the latest real-time news events; wherein the multimodal big model includes a first preset template for extracting the current summary information, and the first preset template is determined according to the format requirements and attribute requirements of the current summary information; Determine the authenticity of the latest real-time news event corresponding to each current summary information according to the verification rules and verification weights corresponding to each news element of the current summary information in the third preset template; According to the authenticity of the latest real-time news event and the current summary information, determining from the latest real-time news event a target latest real-time news event corresponding to the historical request information and the current summary information of the target latest real-time news event; The historical summary information, the historical response and the current summary information of the target latest real-time news event are input into the multimodal large model, and the fusion information for the historical request information is output under the action of the fourth preset template.

[0013] In combination with the first aspect, an embodiment of the present invention provides an eighth possible implementation of the first aspect, wherein the step of inputting the current request information for the real-time news event, the fusion information for the historical request information, and the historical chat record content into the multimodal big model, and outputting a response to the current request information includes: The fused information and the historical chat record content are semantically recognized, and a response to the current request information is determined under the action of a multimodal large model.

[0014] In a second aspect, an embodiment of the present invention further provides a chat robot data processing device based on a large model, comprising: A calling module, in response to current request information of a member in a chat group for a real-time news event, calling a chat robot; wherein the chat group includes at least one of the chat robots; A determination module, which matches the real-time news event with the chat records of the chat group through the chat robot to determine the historical chat record content associated with the real-time news event; wherein the chat robot includes a multimodal large model; A first response module, if the historical chat record content does not contain historical request information for the real-time news event, inputs the historical chat record content and the current request information into the multimodal macro model, and outputs a response to the current request information; wherein the historical request information is associated with the current request information; A fusion module, if the historical chat record content contains historical request information for real-time news events, then respectively determines the historical summary information and historical responses corresponding to the historical request information, and the current summary information and current responses corresponding to the historical request information, and inputs them into the multimodal large model to output fusion information for the historical request information; The second response module inputs the current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content into the multimodal large model, and outputs a response to the current request information.

[0015] The embodiment of the present invention provides a chat robot data processing method and device based on a large model, including a chat robot with a multimodal large language model having the ability to analyze and summarize the target chat record content in real time, and the ability to identify and parse the target chat record content and output a response; the embodiment of the present invention calls the chat robot according to the request of the real-time news event, finds the related historical records of the real-time news event from the chat record, and then checks from the related historical records whether there are any questions about the real-time news event; if not, the multimodal model outputs a response based on the related historical records and the current question; if so, the historical summary response information of the historical questions and the current summary response information of the historical questions are fused, and then the fused information and the historical chat records are input into the model to output the response to the current question; by combining the corresponding fused information of the historical questions about the real-time news event in the chat record, the accuracy of the current question output by the large model can be made higher.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description and the drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flow chart of a chat robot data processing method based on a large model provided by an embodiment of the present invention; Figure 2 A schematic diagram of functional modules of a chat robot data processing device based on a large model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] At present, the reliability of chatbots using multimodal large models in responding to real-time news events is not high. This is because large models generally ignore the real-time nature of their responses to real-time news events, which leads to lags and errors in the response results. For example, how many gold medals did country X win in a certain sports game? This request information is real-time during the sports game, and the number of gold medals changes.

[0022] Based on this, a chat robot data processing method and device based on a large model provided by an embodiment of the present invention can combine the historical request information about the real-time news event in the chat record to obtain an accurate response to the current request information of the real-time news event.

[0023] To facilitate understanding of this embodiment, a chat robot data processing method based on a large model disclosed in an embodiment of the present invention is first introduced in detail. This method can be applied to intelligent devices such as controllers, host computers, servers, and touch-screen interactive terminals.

[0024] Figure 1 A flow chart of a chat robot data processing method based on a large model provided by an embodiment of the present invention; Reference Figure 1 , the method mainly comprises the following steps: Step S102, in response to the current request information of the members in the chat group for the real-time news event, calling the chat robot.

[0025] Among them, the chat group is used for at least three group members to communicate and chat, and each chat group includes at least one chat robot; the request information in the embodiment of the present invention can be understood as a request statement for a real-time news event; the request statement can include multiple modal forms, such as voice, text, video, etc.; after each modal request statement is converted into text, the statement includes a first text information for locating the real-time news event, and a second text information for representing a corresponding request for the text information, so that at least one real-time news event can be retrieved and captured based on the request information. As an optional embodiment, based on the reception of the current request information, the operation of the chat robot can be triggered.

[0026] Step S104, matching the real-time news event with the chat records of the chat group through the chat robot to determine the historical chat record content associated with the real-time news event.

[0027] Among them, the chat robot includes a multimodal large model; the multimodal large model has the ability to perform specific semantic recognition and matching of the real-time news event corresponding to the aforementioned current request information with the chat history of the chat group, and then filter out historical chat content related to the real-time news event; the real-time news event may include at least one.

[0028] Step S106: If the historical chat record content does not contain historical request information for the real-time news event, the historical chat record content and the current request information are input into the multimodal large model, and a response to the current request information is output; Here, the historical chat content associated with the real-time news event is further screened to determine whether the historical chat content contains historical request information for the real-time news event, and the historical request information is associated with the current request information; for example, the historical request information and the current request information are the same request and the same question; if the historical chat record content does not contain the historical request information associated with the real-time news event and the current request information, the response to the current request information is directly determined based on the historical chat record content related to the real-time news event content; It should be noted that the multimodal big model in the embodiment of the present invention has the ability to understand multimodal input content such as video, picture, voice, text, etc., and generate corresponding responses, and has the ability to convert multimodal input content such as video, picture, voice, etc. into text.

[0029] Step S108, if the historical chat record content contains historical request information for real-time news events, the historical summary information and historical responses corresponding to the historical request information, as well as the current summary information and current responses corresponding to the historical request information are determined respectively, and input into the multimodal large model to output the fusion information for the historical request information.

[0030] Here, if there is historical request information in the historical chat record content that is associated with the real-time news event and the current request information, the historical information such as the historical summary and historical response of the historical request information, as well as the real-time news event that can be retrieved and captured at the current moment corresponding to the historical request information, the current summary and current response of such real-time news event at the current moment are used to generate fusion information under the action of the multimodal large model. The fusion information for the historical request information can help the current request information have a more accurate response.

[0031] Step S110, input the current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content into the multimodal big model, and output the response to the current request information.

[0032] Here, an embodiment of the present invention adopts fusion information of historical request information for real-time news events, historical chat record content associated with real-time news events, and summary information corresponding to at least one real-time news event captured based on current request information as input of a multimodal big model, and obtains a response to the current request information through semantic recognition and analysis of the multimodal big model.

[0033] It is understandable that real-time news events may include text, pictures, videos, audio and other forms; the current request information may include questions, summaries and other requests for real-time news events; the multimodal large model is pre-trained and can realize the cross-modal attention mechanism to fuse information of different modalities, and can also understand the specific context of the news event through in-context learning (ICL), and can also use the model's zero-sample transfer capability to make reasonable inferences even if it has never seen similar events.

[0034] In a preferred embodiment of actual application, a chat robot including a multimodal large language model has the ability to analyze and summarize the target chat record content in real time, and the ability to identify and parse the target chat record content and output a response; the embodiment of the present invention calls the chat robot according to the request of the real-time news event, finds the associated historical records of the real-time news event from the chat record, and then checks from the associated historical records whether there are any questions about the real-time news event; if not, the multimodal model outputs a response based on the associated historical records and the current question; if so, the historical summary response information of the historical questions and the current summary response information of the historical questions are fused, and then the fused information and the historical chat records are input into the model to output the response to the current question; by combining the corresponding fused information of the historical questions about the real-time news event in the chat record, the accuracy of the current question output by the large model can be made higher.

[0035] In some embodiments, step S102 of calling a chatbot includes: Step 1.1), in response to a member in the chat group inputting a preset target word for the chat robot, the working state of the chat robot is triggered.

[0036] Here, the same chat group may include at least one member triggering a chat robot at the same time; the preset target word may be the name or nickname of the chat robot to be triggered; wherein the chat robot may be switched to a working state by inputting the preset target word.

[0037] Step 1.2), in response to the current request information input by the group members in the chat group, control the chat robot to determine at least one corresponding real-time news event and summary information corresponding to each real-time news event according to the current request information.

[0038] The preset target word and the current request information can be input through a graphical user interface, which can be understood as an interface integrated in the above-mentioned smart device for facilitating user interaction; In actual applications, the chatbot can crawl corresponding real-time news events from third-party websites based on current request information, and extract corresponding summary information from each real-time news event.

[0039] Furthermore, the chatbot includes a multimodal large model. Step S104 can be implemented by the following steps to ensure that the historical chat record content associated with the real-time news event can be obtained more accurately, specifically including: Step 2.1), extract keywords from real-time news events through the chatbot.

[0040] Here, the chatbot including the multimodal large model has the ability to identify and analyze real-time news events by segmenting the real-time news events and extracting the corresponding keywords (removing prepositions, auxiliary words, conjunctions, etc. that have no practical meaning).

[0041] Step 2.2), based on the extracted keywords, search the chat records of the chat group to determine whether there is any historical chat record content that matches the real-time news event.

[0042] Here, keywords of real-time news events are used to find matching historical chat record content.

[0043] As an optional embodiment, step S104 may also be implemented by the following steps to ensure that the historical chat record content associated with the real-time news event can be more accurately obtained, specifically including: In step 3.1), the chat records of the chat group are classified according to preset target words to obtain a first chat group for representing the chat between the group members and the chat robot and a second chat group for representing the chat between the group members.

[0044] Here, the chat records including the preset target words can be classified as the first chat group, and the rest are defaulted to chats between group members, that is, classified as the second chat group.

[0045] Step 3.2), extract keywords from real-time news events through the chatbot.

[0046] Keywords are obtained from real-time news events in a similar manner to the above-mentioned embodiment, and the specific implementation method is not repeated here.

[0047] Step 3.3), based on the extracted keywords, matching chat record contents are searched from the first chat group and the second chat group respectively, and the matching degree of each matching chat record content is determined.

[0048] Here, when chat record contents matching the keyword are found from the first chat group and the second chat group, each matching chat record content is assigned a corresponding matching score according to its matching degree with the keyword. For example, the matching score of matching chat record content 1 from the first chat group is 50, the matching score of matching chat record content 2 from the second chat group is 80, and so on.

[0049] Step 3.4), assign corresponding weights to the first chat group and the second chat group according to the number of matching chat record contents respectively existing in the first chat group and the second chat group.

[0050] For example, if there are 7 matching chat record contents in the first chat group and 3 matching chat record contents in the second chat group, the weight assigned to the first chat group is 0.7 and the weight assigned to the second chat group is 0.3.

[0051] Step 3.5), based on the matching degree of each matching chat record content and the corresponding weight of the chat group to which each matching chat record content belongs, determine the relevance of each matching chat record content to the real-time news event.

[0052] Here, the relevance of each matching chat record content to the real-time news event is determined based on the product of the matching score of each matching chat record content and the weight of the chat group to which it belongs. For example, the relevance of matching chat record content 1 to the real-time news event is 50*0.7=35, and the relevance of matching chat record content 2 to the real-time news event is 80*0.3=24.

[0053] In step 3.6), by comparing the correlation with the preset correlation threshold, it is determined whether there is any historical chat record content that matches the real-time news event.

[0054] Here, if the correlation exceeds the preset correlation threshold, the matching chat record content is identified as the historical chat record content matching the real-time news event. Otherwise, if there is no matching chat record content with a correlation exceeding the preset correlation threshold, there is no historical chat record content.

[0055] As another optional embodiment, step S104 may also be implemented by the following steps to ensure that the historical chat record content associated with the real-time news event can be more comprehensively acquired, specifically including: Step 4.1), the chat robot performs semantic recognition on the chat records to determine the topic corresponding to the content of each chat record.

[0056] According to the aforementioned embodiments, the chat robot can perform semantic recognition and analysis on the chat records to obtain the subject corresponding to each chat record content. The subject can be understood as the scenario, situation, etc. represented by the chat record content that can represent the intention of the chat record content.

[0057] Step 4.2), based on the matching degree of each chat record content corresponding to the topic, determine the relevance of each chat record content.

[0058] Among them, each chat record content corresponds to a topic, and the relevance of the chat record content corresponding to the topic can be known through the matching degree between the topics; for example, the matching degree of topic 1 of chat record content 1 and topic 2 of chat record content 2 is 60%, then it can be considered that the relevance of chat record content 1 and chat record content 2 is 60%.

[0059] Step 4.3), extract keywords from real-time news events through the chatbot.

[0060] This is similar to the keyword extraction step in the previous embodiment and will not be repeated here.

[0061] In step 4.4), the extracted keywords are searched from each chat record of the chat group to determine the first chat record.

[0062] Here, the keyword is matched with each chat record, and the chat record in which the keyword is found is identified as the first chat record.

[0063] Step 4.5), based on the second chat record whose relevance to the first chat record meets the preset relevance threshold and the first chat record, jointly determine the historical chat record content that matches the real-time news event.

[0064] It can be known that, based on the matching degree of the topic, the first chat record can obtain the relevance of the first chat record to other chat records, and other chat records with a relevance higher than a preset relevance threshold can be identified as the second chat record; based on the first chat record and the second chat record, historical chat record content matching the real-time news event is jointly generated.

[0065] A corresponding method may be selected to execute step S104 according to actual application business requirements.

[0066] In some embodiments, on the basis that the historical chat record content includes the real-time news event, it can be further determined whether the historical chat record content includes a historical request also related to the current request for the real-time news event. Exemplarily, steps S106-S108 can be implemented by the following steps, including: Step 5.1), determine whether the historical chat record content contains historical request information for real-time news events.

[0067] Here, the historical request information is associated with the current request information, and the keywords extracted from the current request information can be used to query whether there is matching content in the historical chat record content; Step 5.2), if not included, perform the following steps: a. Control the real-time search engine to retrieve at least one real-time news event corresponding to the current request information.

[0068] Here, the real-time search engine can crawl at least one real-time news event corresponding to the first text information of the current request information from at least one third-party website.

[0069] b. Extract the corresponding key information from each real-time news event and input the key information into the multimodal large model to determine the summary information corresponding to each real-time news event.

[0070] Among them, the multimodal large model includes a first preset template for extracting summary information, and the first preset template is determined according to the format requirements and attribute requirements of the summary information. The multimodal large model converts the summary information into text form through the first preset template to extract text. Among them, the format requirements and attribute requirements satisfied by the expected summary information are pre-set; the format requirements can be understood as the arrangement of the text in the summary information, and the attribute requirements can be understood as the content information of the text in the summary information, and the attributes corresponding to the content information; and the first preset template is determined according to the format requirements and attribute requirements of the summary information; the first preset template is used to guide the multimodal large model to further extract summary information from the key information of each real-time news event; here, the key information can be understood as the information that can reflect the main content of the real-time news event that is initially captured from each real-time news event, usually including at least one complete news paragraph in the real-time news event.

[0071] c. Determine the authenticity of the real-time news event corresponding to each summary information according to the verification rules and verification weights corresponding to each news element of the summary information in the third preset template.

[0072] It should be noted that the current real-time news events are captured only because they are related to the requested information, and the authenticity of such real-time news events themselves cannot be guaranteed. The embodiment of the present invention verifies the news elements corresponding to each real-time news event through a third preset template to determine the authenticity of each real-time news event.

[0073] Among them, each news event includes at most six news elements, such as time, people, place, cause, process, and result; the third preset template is pre-configured with verification rules for each news element, as well as the verification weight corresponding to each news element, that is, whether each news element is verified or not plays a different role in the authenticity of the real-time news event.

[0074] d. According to the authenticity and summary information of the real-time news event, determine the target real-time news event corresponding to the current request information from the real-time news events.

[0075] It can be understood that, by combining the verification of the authenticity of each real-time news event in the aforementioned embodiment steps and the summary information extracted from each real-time news event, the target real-time news event with the highest match to the requested information is determined; in other words, the target real-time news event is generally a real-time news event that meets the requested information and has a higher degree of authenticity.

[0076] e. Input the current request information, historical chat record content and summary information of the target real-time news event into the multimodal large model, and output a response to the current request information under the action of the second preset template.

[0077] The multimodal large model also includes a second preset template, which can fuse the historical chat record content and the summary information of the target real-time news event, so that the multimodal large model can perform more accurate recognition and analysis based on the fused information, and can respond to the current request information input for the target real-time news event and output reply information for the request information.

[0078] Step 5.3), if included, execute the following steps: x. Obtain historical summary information and historical responses corresponding to historical request information from historical chat record content through the chat robot.

[0079] It should be noted that the context chat record content of the historical request information is located from the historical chat record content based on the keywords of the historical request information, and the historical summary information of the real-time news event corresponding to the historical request information at that time and the historical response are obtained from the context chat record content.

[0080] y. Input the current summary of at least one latest real-time news event corresponding to the historical summary information, historical response, and historical request information at the current moment into the multimodal large model, and output the fusion information for the historical request information, which can be specifically implemented by the following steps: y1. Control the real-time search engine to retrieve and extract at least one latest real-time news event corresponding to the historical request information at the current moment.

[0081] Here, the real-time search engine can crawl at least one real-time news event corresponding to the first text information of the historical request information from at least one third-party website.

[0082] y2. Extract the corresponding key information from each latest real-time news event, and input the key information into the multimodal large model to determine the current summary information corresponding to each latest real-time news event.

[0083] Among them, the multimodal large model includes a first preset template for extracting summary information, and the first preset template is determined according to the format requirements and attribute requirements of the summary information. Among them, the format requirements and attribute requirements satisfied by the expected summary information are preset; the format requirements can be understood as the arrangement of the text in the summary information, and the attribute requirements can be understood as the content information of the text in the summary information, and the attributes corresponding to the content information; and the first preset template is determined according to the format requirements and attribute requirements of the summary information; the first preset template is used to guide the multimodal large model to further extract summary information from the key information of each real-time news event; here, the key information can be understood as the information that can reflect the main content of the real-time news event that is initially captured from each real-time news event, usually including at least one complete news paragraph in the real-time news event.

[0084] y3. Determine the authenticity of the latest real-time news event corresponding to each current summary information according to the verification rules and verification weights corresponding to each news element of the current summary information in the third preset template.

[0085] It should be noted that the current real-time news events are captured only because they are related to the requested information, and the authenticity of such real-time news events themselves cannot be guaranteed. The embodiment of the present invention verifies the news elements corresponding to each real-time news event through a third preset template to determine the authenticity of each real-time news event.

[0086] Among them, each news event includes at most six news elements, such as time, people, place, cause, process, and result; the third preset template is pre-configured with verification rules for each news element, as well as the verification weight corresponding to each news element, that is, whether each news element is verified or not plays a different role in the authenticity of the real-time news event.

[0087] y4. According to the authenticity and current summary information of the latest real-time news event, determine the target latest real-time news event corresponding to the historical request information and the current summary information of the target latest real-time news event from the latest real-time news event.

[0088] It can be understood that, by combining the verification of the authenticity of each latest real-time news event in the aforementioned embodiment steps and the summary information extracted from each latest real-time news event, the target latest real-time news event with the highest match to the historical request information is determined; in other words, the target latest real-time news event is generally the latest real-time news event that meets the historical request information and has a higher degree of authenticity.

[0089] y5. Input the historical summary information, historical responses and current summary information of the target latest real-time news event into the multimodal large model, and output the fusion information for the historical request information under the action of the fourth preset template.

[0090] The multimodal large model also includes a fourth preset template, which can integrate the historical summary information, historical responses, and current summary information of the target latest real-time news event corresponding to the historical request information.

[0091] In some embodiments, step S110 can more accurately respond to the current request information under the effect of the fusion information corresponding to the historical request associated with the current request information and the historical chat record content matching the real-time news event corresponding to the current request information; specifically, it includes: In step 6.1), semantic recognition is performed on the fused information and historical chat record content, and the response to the current request information is determined under the action of the multimodal large model.

[0092] In some embodiments, Figure 2 As shown, an embodiment of the present invention provides a chat robot data processing device based on a large model, comprising: A calling module, in response to current request information of a member in a chat group for a real-time news event, calling a chat robot; wherein the chat group includes at least one of the chat robots; A determination module, which matches the real-time news event with the chat records of the chat group through the chat robot to determine the historical chat record content associated with the real-time news event; wherein the chat robot includes a multimodal large model; A first response module, if the historical chat record content does not contain historical request information for the real-time news event, inputs the historical chat record content and the current request information into the multimodal macro model, and outputs a response to the current request information; wherein the historical request information is associated with the current request information; A fusion module, if the historical chat record content contains historical request information for real-time news events, then respectively determines the historical summary information and historical responses corresponding to the historical request information, and the current summary information and current responses corresponding to the historical request information, and inputs them into the multimodal large model to output fusion information for the historical request information; The second response module inputs the current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content into the multimodal large model, and outputs a response to the current request information.

[0093] Furthermore, the calling module is also specifically used to trigger the working state of the chat robot in response to a preset target word input operation of the chat robot by a member in the chat group; and in response to current request information input by a member in the chat group, control the chat robot to determine at least one corresponding real-time news event according to the current request information, as well as summary information corresponding to each real-time news event.

[0094] Furthermore, the determination module is also specifically used to extract keywords from the real-time news event through the chat robot; based on the extracted keywords, search from the chat records of the chat group to determine whether there is historical chat record content that matches the real-time news event.

[0095] Furthermore, the determination module is also specifically used to classify the chat records of the chat group according to the preset target words to obtain a first chat group for characterizing the chat between the group members and the chat robot and a second chat group for characterizing the chat between the group members; extract keywords from the real-time news event through the chat robot; search for matching chat record content from the first chat group and the second chat group based on the extracted keywords, and determine the matching degree of each matching chat record content; assign corresponding weights to the first chat group and the second chat group according to the number of matching chat record content respectively existing in the first chat group and the second chat group; determine the correlation degree of each matching chat record content with the real-time news event based on the matching degree of each matching chat record content and the corresponding weight of the chat group to which each matching chat record content belongs; and judge whether there is historical chat record content matching the real-time news event by comparing the correlation degree with a preset correlation threshold.

[0096] Furthermore, the determination module is specifically used to perform semantic recognition of the chat records by the chat robot to determine the topic corresponding to the content of each chat record; determine the relevance of each chat record content based on the matching degree of the topics corresponding to the contents of each chat record; extract keywords from the real-time news event by the chat robot; search for the extracted keywords from each chat record of the chat group to determine the first chat record; and jointly determine the historical chat record content that matches the real-time news event based on the second chat record and the first chat record whose relevance to the first chat record meets a preset relevance threshold.

[0097] Furthermore, the first response module is also specifically used to determine whether the content of the historical chat record contains historical request information for real-time news events; if not, the following steps are performed: controlling the real-time search engine to retrieve at least one real-time news event corresponding to the current request information; extracting corresponding key information from each real-time news event, and inputting the key information into the multimodal large model to determine the summary information corresponding to each real-time news event; wherein the multimodal large model includes a first preset template for extracting the summary information, and the first preset template is determined according to the format requirements and attribute requirements of the summary information; according to the verification rules and verification weights corresponding to each news element of the summary information in the third preset template, the authenticity of the real-time news event corresponding to each summary information is determined; according to the authenticity of the real-time news event and the summary information, the target real-time news event corresponding to the current request information is determined from the real-time news event; the current request information, the historical chat record content and the summary information of the target real-time news event are input into the multimodal large model, and a response to the current request information is output under the action of the second preset template.

[0098] Furthermore, the fusion module is also specifically used to determine whether the historical chat record content contains historical request information for real-time news events; if contained, the following steps are executed: obtaining historical summary information and historical responses corresponding to the historical request information from the historical chat record content through the chat robot; inputting the current summary of the historical summary information, the historical response, and the current summary of at least one latest real-time news event corresponding to the historical request information at the current moment into the multimodal large model, and outputting fusion information for the historical request information.

[0099] Furthermore, the fusion module is also specifically used to control the real-time search engine to retrieve and extract at least one latest real-time news event corresponding to the historical request information at the current moment; extract corresponding key information from each of the latest real-time news events, and input the key information into the multimodal big model to determine the current summary information corresponding to each of the latest real-time news events; wherein the multimodal big model includes a first preset template for extracting the current summary information, and the first preset template is determined according to the format requirements and attribute requirements of the current summary information; according to the verification rules and verification weights corresponding to each news element of the current summary information in the third preset template, the authenticity of the latest real-time news event corresponding to each of the current summary information is determined; according to the authenticity of the latest real-time news event and the current summary information, the target latest real-time news event corresponding to the historical request information and the current summary information of the target latest real-time news event are determined from the latest real-time news event; the historical summary information, the historical response and the current summary information of the target latest real-time news event are input into the multimodal big model, and the fusion information for the historical request information is output under the action of the fourth preset template.

[0100] Furthermore, the second response module is also specifically used to perform semantic recognition on the fusion information and the historical chat record content, and determine the response to the current request information under the action of the multimodal large model.

[0101] It can be understood that the specific operation methods of each functional module in this embodiment can refer to the detailed description of the corresponding steps in the above method embodiment, and will not be repeated here.

[0102] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0103] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0104] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0105] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the aforementioned embodiments, those of ordinary skill in the art should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the aforementioned embodiments within the technical scope disclosed by the present invention, or can easily conceive of changes, or make equivalent replacements for some of the technical features therein. Such modifications, changes or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the protection scope of the present invention.

Claims

1. A chat robot data processing method based on a large model, characterized in that: include: In response to current request information of a member in a chat group for a real-time news event, calling a chat robot; wherein the chat group includes at least one of the chat robots; The chat robot matches the real-time news event with the chat records of the chat group to determine the historical chat record content associated with the real-time news event; wherein the chat robot includes a multimodal large model; If the historical chat record content does not contain historical request information for the real-time news event, the historical chat record content and the current request information are input into the multimodal macro model, and a response to the current request information is output; wherein the historical request information is associated with the current request information; If the historical chat record content contains historical request information for real-time news events, the historical summary information and historical responses corresponding to the historical request information, as well as the current summary information and current responses corresponding to the historical request information are determined respectively, and input into the multimodal large model to output fusion information for the historical request information; The current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content are input into the multimodal big model, and a response to the current request information is output.

2. The method according to claim 1, characterized in that In response to a current request for information about a real-time news event from a member of the chat group, the step of invoking the chat robot comprises: In response to a preset target word input operation of a chat robot by a member of the chat group, triggering a working state of the chat robot; In response to current request information input by a member of the chat group, the chat robot is controlled to determine at least one corresponding real-time news event and summary information corresponding to each real-time news event according to the current request information.

3. The method according to claim 1, characterized in that The step of matching the real-time news event with the chat records of the chat group by the chat robot to determine the historical chat record content associated with the real-time news event includes: Extracting keywords from the real-time news event by using the chat robot; Based on the extracted keywords, the chat records of the chat group are searched to determine whether there is historical chat record content that matches the real-time news event.

4. The method according to claim 2, characterized in that: The step of matching the real-time news event with the chat records of the chat group by the chat robot to determine the historical chat record content associated with the real-time news event includes: Classifying the chat records of the chat group according to the preset target words to obtain a first chat group for representing the chat between the group member and the chat robot and a second chat group for representing the chat between the group member and the group member; Extracting keywords from the real-time news event by using the chat robot; searching for matching chat record content from the first chat group and the second chat group respectively based on the extracted keywords, and determining a matching degree of each matching chat record content; Allocating corresponding weights of the first chat group and the second chat group according to the number of the matching chat record contents respectively existing in the first chat group and the second chat group; Determine the relevance of each of the matched chat record contents to the real-time news event based on the matching degree of each of the matched chat record contents and the corresponding weight of the chat group to which each of the matched chat record contents belongs; By comparing the correlation with a preset correlation threshold, it is determined whether there is historical chat record content that matches the real-time news event.

5. The method according to claim 1, characterized in that: The step of matching the real-time news event with the chat records of the chat group by the chat robot to determine the historical chat record content associated with the real-time news event includes: The chat robot performs semantic recognition on the chat records to determine the subject corresponding to each chat record content; Determining the relevance of each chat record content based on the matching degree of each chat record content corresponding to the topic; Extracting keywords from the real-time news event by using the chat robot; Searching the extracted keywords from each chat record of the chat group to determine a first chat record; The historical chat record content matching the real-time news event is jointly determined based on the second chat record and the first chat record whose relevance to the first chat record meets a preset relevance threshold.

6. The method according to claim 1, characterized in that If the historical chat record content does not contain historical request information for real-time news events, the steps of inputting the historical chat record content and the current request information into the multimodal macro model and outputting a response to the current request information include: Determining whether the historical chat record content contains historical request information for real-time news events; If not, perform the following steps: Controlling the real-time search engine to retrieve at least one real-time news event corresponding to the current request information; Extracting corresponding key information from each of the real-time news events, and inputting the key information into the multimodal big model, and determining summary information corresponding to each of the real-time news events; wherein the multimodal big model includes a first preset template for extracting the summary information, and the first preset template is determined according to the format requirements and attribute requirements of the summary information; Determining the authenticity of the real-time news event corresponding to each of the summary information according to the verification rules and verification weights corresponding to each news element of the summary information in the third preset template; Determining a target real-time news event corresponding to the current request information from the real-time news events according to the authenticity of the real-time news event and the summary information; The current request information, the historical chat record content and the summary information of the target real-time news event are input into the multimodal large model, and a response to the current request information is output under the action of a second preset template.

7. The method according to claim 1, characterized in that If the historical chat record content contains historical request information for real-time news events, the steps of respectively determining historical summary information and historical responses corresponding to the historical request information, and current summary information and current responses corresponding to the historical request information, and inputting them into a multimodal large model to output fusion information for the historical request information include: Determining whether the historical chat record content contains historical request information for real-time news events; If included, perform the following steps: Acquiring, by the chat robot, historical summary information and historical responses corresponding to the historical request information from the historical chat record content; The historical summary information, the historical response, and the current summary of at least one latest real-time news event corresponding to the historical request information at the current moment are input into a multimodal large model, and fusion information for the historical request information is output.

8. The method according to claim 7, characterized in that The steps of inputting the historical summary information, the historical response, and the current summary of at least one latest real-time news event corresponding to the historical request information at the current moment into the multimodal macro model, and outputting the fusion information for the historical request information include: Control the real-time search engine to retrieve and extract at least one latest real-time news event corresponding to the historical request information at the current moment; Extracting corresponding key information from each of the latest real-time news events, and inputting the key information into the multimodal big model, and determining the current summary information corresponding to each of the latest real-time news events; wherein the multimodal big model includes a first preset template for extracting the current summary information, and the first preset template is determined according to the format requirements and attribute requirements of the current summary information; Determine the authenticity of the latest real-time news event corresponding to each current summary information according to the verification rules and verification weights corresponding to each news element of the current summary information in the third preset template; According to the authenticity of the latest real-time news event and the current summary information, determining from the latest real-time news event a target latest real-time news event corresponding to the historical request information and the current summary information of the target latest real-time news event; The historical summary information, the historical response and the current summary information of the target latest real-time news event are input into the multimodal large model, and the fusion information for the historical request information is output under the action of the fourth preset template.

9. The method according to claim 1, characterized in that: The step of inputting the current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content into the multimodal macro model, and outputting a response to the current request information comprises: The fused information and the historical chat record content are semantically recognized, and a response to the current request information is determined under the action of a multimodal large model.

10. A chat robot data processing device based on a large model, characterized in that: include: A calling module, in response to current request information of a member in a chat group for a real-time news event, calls a chat robot; wherein the chat group includes at least one of the chat robots; A determination module, which matches the real-time news event with the chat records of the chat group through the chat robot to determine the historical chat record content associated with the real-time news event; wherein the chat robot includes a multimodal large model; A first response module, if the historical chat record content does not contain historical request information for the real-time news event, inputs the historical chat record content and the current request information into the multimodal macro model, and outputs a response to the current request information; wherein the historical request information is associated with the current request information; A fusion module, if the historical chat record content contains historical request information for real-time news events, then respectively determines the historical summary information and historical responses corresponding to the historical request information, and the current summary information and current responses corresponding to the historical request information, and inputs them into the multimodal large model to output fusion information for the historical request information; The second response module inputs the current request information for the real-time news event, the fusion information for the historical request information and the historical chat record content into the multimodal large model, and outputs a response to the current request information.

Images