Multi-theme dialogue processing method and device for real-time news based on large model

Abstract

The invention provides a multi-theme dialogue processing method and device for real-time news based on a large model, and relates to the technical field of dialogue processing of real-time news, and the method comprises the steps: obtaining a first theme corresponding to current request information, a real-time news event and corresponding first summary information; according to the authenticity of each real-time news event and the first summary information, determining a target real-time news event and corresponding current summary information; performing relevance matching on the first theme and a second theme corresponding to each piece of historical request information, and updating a node tree according to a matching result; according to the updated node tree, determining target historical request information of which the association degree with the first theme meets a preset requirement; inputting the current request, the current abstract, the target historical request and the corresponding historical abstract and historical response into a multi-modal large model, and outputting a response for the current request information; the technical problem that in the prior art, the response accuracy of real-time news is not high is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of dialogue processing of real-time news, and in particular to a method and device for multi-topic dialogue processing for real-time news 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] Current multimodal large models generally obtain answers to user questions by pre-setting an updateable knowledge base; however, for real-time news events, such answers may be delayed and thus fail to meet the user's question requirements regarding real-time news events; in addition, if the user initiates a multi-topic dialogue request regarding a real-time news event, the accuracy of the response to the request in the dialogue will be further greatly reduced. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-topic dialogue processing method and device for real-time news based on a large model, so as to alleviate the technical problem of low accuracy of response to real-time news in the prior art.

[0005] In a first aspect, an embodiment of the present invention provides a multi-topic dialogue processing method for real-time news based on a large model, comprising: In response to current request information inputted into the graphical user interface, determining a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event; According to the authenticity of each of the real-time news events and each of the first summary information, determining a target real-time news event corresponding to the current request information and current summary information corresponding to the target real-time news event from the real-time news events; Match the first topic with the second topic corresponding to each historical request information, and update the node tree according to the matching result; wherein the node tree includes at least one node, the node is used to represent the topic corresponding to each request information, and the node tree is used to represent the relevance of each request information for real-time news events; Determine, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements; The current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information are input into a multimodal large 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 performing relevance matching between the first topic and the second topic corresponding to each piece of historical request information, and updating the node tree according to the matching result includes: Matching the first topic with the second topic corresponding to each of the historical request information according to the neural network; Determining, based on the matching result, whether there is a second topic whose relevance to the first topic meets a preset relevance threshold; If so, determine the node corresponding to the second topic whose relevance to the first topic meets the preset relevance threshold from the node tree as the target node, and update the node tree by taking the node corresponding to the first topic as the child node of the target node; If it does not exist, the node tree is updated by taking the node corresponding to the first theme as the root node.

[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 first topic with the second topic corresponding to each of the historical request information according to the neural network includes: Performing word segmentation processing on the first topic and the second topic corresponding to each historical request information respectively through a neural network to determine the word segmentation types of the first topic and the second topic; wherein the word segmentation types include entities and connecting relationship words; Based on the word segmentation type, the first topic and the second topic are respectively matched in terms of relevance.

[0008] In combination with the first aspect, an embodiment of the present invention provides a third possible implementation of the first aspect, wherein, before the step of performing relevance matching between the first topic and the second topic corresponding to each piece of historical request information and updating the node tree according to the matching result, the method further includes: A second topic corresponding to each piece of historical request information is jointly determined according to the historical request information and the historical responses corresponding to the historical request information.

[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 determining, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements includes: According to the updated node tree, determine the new node of the node tree; Based on the new node, target historical request information whose correlation with the current request information meets preset requirements is determined.

[0010] In combination with the first aspect, an embodiment of the present invention provides a fifth possible implementation of the first aspect, wherein the step of determining, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements includes: According to the updated node tree, determine the new node of the node tree; Based on the new node and the historical request information corresponding to each node having a dependent relationship with the new node, target historical request information whose correlation with the current request information meets preset requirements is determined.

[0011] In combination with the first aspect, an embodiment of the present invention provides a sixth possible implementation of the first aspect, wherein, in response to current request information input to a graphical user interface, the step of determining a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event includes: In response to current request information inputted into the graphical user interface, controlling the real-time search engine to retrieve at least one real-time news event corresponding to the current request information; wherein the current request information includes a first request statement for locating at least one real-time news event and a second request statement for representing a corresponding request for the real-time news event; Processing the first request statement and the second request statement to determine a first topic corresponding to the current request information; wherein the first topic is composed of an entity and a connecting relation word; Corresponding key information is extracted from each of the real-time news events, and the key information is input into a multimodal big model to determine the first summary information corresponding to each of the real-time news events; wherein the multimodal big model includes a first preset template for extracting the first summary information, and the first preset template is determined according to the format requirements and attribute requirements of the first summary information.

[0012] In combination with the first aspect, the embodiment of the present invention provides a seventh possible implementation manner of the first aspect, wherein, according to the authenticity of each of the real-time news events and each of the first summary information, the step of determining, from the real-time news events, a target real-time news event corresponding to the current request information, and the current summary information corresponding to the target real-time news event includes: Determine the authenticity of the real-time news event corresponding to each of the first summary information according to the verification rules and verification weights corresponding to each news element of the first summary information in the second preset template of the multimodal large model; If the authenticity meets the requirements, the real-time news event is determined as a news fact, and according to the qualifier in the first summary information, a target real-time news event corresponding to the current request information is determined from the news facts, and the first summary information corresponding to the target real-time news event is used as the current summary information; If the authenticity does not meet the requirements, the real-time news event is determined as a news clue, and the news clue in the real-time news event is removed.

[0013] In combination with the first aspect, an embodiment of the present invention provides an eighth possible implementation manner of the first aspect, wherein the step of inputting the current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information into a multimodal large model, and outputting a response to the current request information includes: The current summary information, the target historical request information, the historical summary information and the historical response of the target real-time news event corresponding to the target historical request information are semantically identified, and the response to the current request information is determined under the action of the multimodal large model.

[0014] In a second aspect, an embodiment of the present invention further provides a multi-topic dialogue processing device for real-time news based on a large model, comprising: A first determination module, in response to current request information inputted into the graphical user interface, determines a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event; a second determination module, which determines, from the real-time news events, a target real-time news event corresponding to the current request information and current summary information corresponding to the target real-time news event, according to the authenticity of each real-time news event and each first summary information; A matching module is used to match the first topic with the second topic corresponding to each historical request information, and update the node tree according to the matching result; wherein the node tree includes at least one node, the node is used to represent the topic corresponding to each request information, and the node tree is used to represent the relevance of each request information for real-time news events; A third determination module determines, based on the updated node tree, target historical request information whose relevance to the first topic meets preset requirements; The response module inputs the current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information into the multimodal large model, and outputs a response to the current request information.

[0015] The embodiment of the present invention provides a multi-topic dialogue processing method and device for real-time news based on a large model, which determines the corresponding first topic, target real-time news event and current summary according to the current request information; determines the second topic with an expected correlation and the historical request corresponding to the second topic according to the first topic; determines the target historical request information that has an association dependency with the current request information based on a node tree that characterizes the correlation dependency between the current request and each request information, and then provides the target historical request information and the corresponding historical summary and response, as well as the summary corresponding to the target real-time news event of the current request, to the multimodal large model so that it outputs an accurate response to the current request information.

[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 multi-topic dialogue processing method for real-time news based on a large model provided by an embodiment of the present invention; Figure 2 A schematic diagram of a node tree provided by an embodiment of the present invention; Figure 3 A schematic diagram of functional modules of a multi-topic dialogue processing device for real-time news 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, multimodal large models generally ignore the real-time nature of 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 the XX Games? This request information is real-time during the XX Games, and the number of gold medals changes.

[0022] Based on this, an embodiment of the present invention provides a multi-topic dialogue processing method and device for real-time news based on a large model, which accurately determines historical requests with dependencies according to the topics, and uses information such as responses corresponding to the historical requests with the highest correlation to determine the precise response to the current request.

[0023] To facilitate understanding of this embodiment, a multi-topic dialogue processing method for real-time news 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 multi-topic dialogue processing method for real-time news 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 current request information input into the graphical user interface, determining a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event.

[0025] Among them, the current request information can be input through a graphical user interface, and the graphical user interface can be understood as an interface integrated in the above-mentioned smart device that is convenient for user interaction; the request information in the embodiment of the present invention can be understood as a request statement for a real-time news event; based on the request information, at least one real-time news event can be retrieved and captured accordingly; the request information can be processed by word segmentation, semantic recognition, etc. to obtain a first topic corresponding to the request information, and the first topic can be understood as the dialogue intention of the current request information; for example, the current request information is how many gold medals country X has in the XX Games, and its corresponding first topic is the number of gold medals in the XX Games of country X; each real-time news event has a corresponding summary information, which is convenient for the multimodal large language model to parse the real-time news event more accurately.

[0026] Step S104, according to the authenticity of each real-time news event and each first summary information, determining from the real-time news events a target real-time news event corresponding to the current request information, and current summary information corresponding to the target real-time news event.

[0027] Here, not every real-time news event corresponding to the current request information is necessarily true and reliable; in order to further improve the response reliability, the embodiment of the present invention further selects the real-time news events corresponding to the current request information, matches the first summary information of each real-time news event with the current request information, and selects the real-time news events with a higher matching degree and authenticity that meet the requirements as target real-time news events to participate in subsequent processing steps.

[0028] Step S106: perform correlation matching between the first topic and the second topic corresponding to each historical request information, and update the node tree according to the matching result.

[0029] Among them, the node tree is used to represent the correlation between various request information for real-time news events; there is a node tree representing the correlation between various request information for real-time news events in advance, and as the current request information is input, the node tree is updated according to the correlation matching degree between the first topic corresponding to the current request information and the second topic corresponding to the historical request information. The updated node tree can represent the correlation between the current request information and various historical request information for real-time news events; the node tree includes at least one, the node tree includes at least one node, and the node is used to represent the topic corresponding to each request information.

[0030] It should be noted that in actual applications, users may input some questions or requests that are not directed to real-time news. The embodiments of the present invention are mainly directed to question requests for real-time news, and ensure the reliability of such responses to real-time news. The request information corresponding to each node in the node tree is directed to real-time news events, but the real-time news events directed to each request information are not necessarily the same, and there may be differences. For example, node 1 corresponds to real-time news events 1-5, and node 2 corresponds to real-time news events 7-8. For example, node 1 is directed to the real-time news event the number of gold medals in the XX Games of Country X, and node 2 is directed to the real-time news event the number of athletes who won gold medals in the XX Games of Country X yesterday.

[0031] Step S108: determining, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements.

[0032] Here, the updated node tree includes the node corresponding to the first topic. The updated node tree can represent the correlation between each request information. The target node with dependent correlation is searched according to the node corresponding to the first topic in the node tree to locate the target historical request information.

[0033] Step S110, input the current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information into the multimodal large model, and output the response to the current request information.

[0034] Here, one current / history request information corresponds to at least one target real-time news event, and each target real-time news event has one current / history summary information, or one current / history request information corresponds to at least one target real-time news event, and each target real-time news event is integrated into one current / history summary information; 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. After the semantic recognition and analysis of the multimodal large model, the response to the current request information can be obtained. It 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.

[0035] In a preferred embodiment of actual application, the first topic, target real-time news event and current summary corresponding to the current request information are determined; a second topic with an expected correlation and the historical request corresponding to the second topic are determined based on the first topic; the target historical request information having correlation dependency with the current request information is determined based on a node tree representing the correlation dependency between the current request and each request information, and then the target historical request information and the corresponding historical summary and response, as well as the summary corresponding to the target real-time news event of the current request are provided to the multimodal large model so that it can output an accurate response to the current request information.

[0036] In some embodiments, step S102 determines the first topic of the currently requested information, the corresponding real-time news event, and the first summary information, which can be specifically implemented by the following steps, including: Step 1.1), in response to current request information input into the graphical user interface, control the real-time search engine to retrieve at least one real-time news event corresponding to the current request information.

[0037] Among them, the current request information includes a first request statement for locating at least one real-time news event and a second request statement for characterizing a corresponding request for the real-time news event; the request statement may include multiple modal forms, such as voice, text, video, etc.; after converting each modal request statement into text, the statement text may include first text information for locating the real-time news event, and second text information for characterizing a corresponding request for the text information; based on the request information, at least one real-time news event can be retrieved accordingly.

[0038] For example, the current request information is how many gold medals country X won in the XX Games, and the corresponding first text information is how many gold medals country X won in the XX Games, and the second text information is how many; then based on the first text information and the second text information, it can be known that the current request information is a question about the Olympic Games.

[0039] Step 1.2), the first request statement and the second request statement are processed to determine the first subject corresponding to the current request information.

[0040] Among them, the first topic is composed of entities and connecting relationship words; the first topic can be obtained as the number of gold medals in the XX Games of Country X by first converting the first request statement and the second request statement of each modality into the first text information and the second text information, and then deleting the auxiliary words and prepositions that have no practical meaning in the first text information and the second text information.

[0041] In step 1.3), corresponding key information is extracted from each real-time news event, and the key information is input into the multimodal large model to determine the first summary information corresponding to each real-time news event.

[0042] Among them, the multimodal large model includes a first preset template for extracting the first summary information, wherein the format requirements and attribute requirements that the expected first summary information meets are preset; the format requirements can be understood as the arrangement of the text in the first summary information, and the attribute requirements can be understood as the content information of the text in the first 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 first summary information; the first preset template is used to guide the multimodal large model to further extract the first summary information from the key information of each real-time news event; here, the key information can be understood as the information that is initially captured from each real-time news event and can reflect the main content of the real-time news event, usually including at least one complete news paragraph in the real-time news event.

[0043] In some embodiments, in order to further ensure the accuracy of the response, the real-time news events corresponding to the current request information are screened to determine the target real-time news event with higher authenticity and more matching with the current request information, and the current summary information of the target real-time news event is used to participate in the application of subsequent steps; here, the step of determining the target real-time news event in step S104 can be implemented by the following steps, including: Step 2.1), according to the verification rules and verification weights corresponding to each news element of the first summary information in the second preset template of the multimodal large model, the authenticity of the real-time news event corresponding to each first summary information is determined, specifically including the following steps: In step 2.1.1), each news element in each first summary information is compared with the verification rules corresponding to each news element in the second preset template to obtain the completeness of each news element.

[0044] Among them, the second preset template configures corresponding verification rules for each news element according to the criticality of each news element; for example, the more critical the news element is, the higher the completeness requirement is, and the lower the probability of 100% matching is; generally speaking, in order of criticality from high to low, the news elements can be sorted as follows: time, result, person, place, process and cause; that is, the completeness requirement of the news element time is relatively high. Taking the first summary information "The tenth competition day of the A XX Games-August 5-will produce-Y-pieces-XX Games gold medals" as an example, it can be determined whether there is a time news element in the first summary information, and the completeness of the text information representing the time in the first summary information can be verified. The time includes year, month, day, hour, minute, and second, and the first summary information contains month and day, so the completeness of the time news element can be verified.

[0045] In step 2.1.2), the authenticity score of each first summary information is determined based on the verification weight of each news element and the completeness of each news element.

[0046] Based on the above-mentioned embodiment, the verification weight corresponding to the more critical news elements is also higher; the product of the completeness of each news element in each first summary information and its verification weight is added together to obtain the authenticity score of each first summary information.

[0047] In step 2.1.3), the authenticity score of each first summary information is compared with a preset score threshold to determine the authenticity of the real-time news event corresponding to each first summary information.

[0048] Here, if the authenticity score corresponding to the first summary information of the real-time news event is higher than the preset score threshold, the real-time news event is deemed to meet the authenticity requirements, and such news event is a news fact; otherwise, the authenticity of the real-time news event does not meet the requirements, and such news event is classified as a news clue, which lacks the completeness of news elements and is likely to be false.

[0049] In step 2.2), if the authenticity meets the requirements, the real-time news event is determined as a news fact, and based on the qualifier in the first summary information, the target real-time news event corresponding to the current request information is determined from the news facts, and the first summary information corresponding to the target real-time news event is used as the current summary information.

[0050] Exemplarily, according to the category of the qualifier in the first summary information, the real domain of the subject and / or object in the first summary information defined by the qualifier is determined; based on the real domain of the first summary information, the target real-time news event matching the requested information is determined from the news facts.

[0051] Since the first request statement in the request information may point to multiple news types or multiple application scenario types at the same time, the embodiment of the present invention can determine the target real-time news event corresponding to the target news type / target scenario / real field through the category of the qualifier in the first summary information; Taking the aforementioned first summary information as an example, the real-time news event may include other activities or competitions where gold medals were won. The gold medals won in the XX Games to which it belongs can be determined by the qualifier XX Games before the name of the gold medal in the first summary information "A session of the XX Games, the tenth competition day - August 5 - will produce - 19 - XX Games gold medals" rather than the gold medals in the game scene, that is, the first summary information of other real-time news events where the gold medals won in the game scene are eliminated; For another example, "Xiao Li" may exist in multiple fields, such as entertainer Xiao Li, athlete Xiao Li, etc. If the requested information is "how many gold medals did Xiao Li win in XX sports games", then based on the first summary information "Xiao Li, a diving athlete, won a gold medal in XX sports games", it can be known from the qualifiers before the subject and object that Xiao Li, who participated in the XX sports games and won the gold medal, is a diving athlete. Therefore, the first summary information of news events about other professions such as entertainers who also won gold medals but whose winners are the same name as Xiao Li will be eliminated.

[0052] In step 2.3), if the authenticity does not meet the requirements, the real-time news event is determined as a news clue, and the news clue in the real-time news event is removed.

[0053] Similarly, news clues that do not meet the authenticity requirements will be eliminated.

[0054] In some embodiments, step S106 determines the target historical request with the highest dependency on the current request based on the correlation between the first topic of the current request and the second topic of the historical request, thereby ensuring the accuracy of the response, and specifically includes: In step 3.1), the first topic is matched with the second topic corresponding to each historical request information according to the neural network.

[0055] Here, according to the general matching association model formed by the neural network, the first subject of the current request information and the second subject of each historical request information can be segmented and matched in terms of association.

[0056] Exemplarily, the first topic and the second topic corresponding to each historical request information are respectively segmented through a neural network to determine the segmentation types of the first topic and the second topic; wherein the segmentation types include entities and connecting relationship words; based on the segmentation types, the first topic and the second topic are respectively matched for relevance.

[0057] Here, it can be understood that the first topic and the second topic are segmented according to entities and connecting relationship words respectively, and then matched according to the entity relevance of the first topic and the second topic, and / or matched according to the connecting relationship word relevance.

[0058] For example, the entity of the first topic is the number of gold medals won by X country in the XX Games, the entity of the second topic A is the number of gold medals won by Xiao Li in the XX Games, and the entity of the second topic B is the number of gold medals won by Xiao Li in diving. The first topic is matched with each second topic respectively.

[0059] Step 3.2), based on the matching results, determine whether there is a second topic whose relevance to the first topic meets a preset relevance threshold.

[0060] As an optional embodiment, the correlation between the first topic and the second topic can be determined based on the ratio of the number of matched words to the total number of words; for example, the correlation between the first topic and a second topic A is 75%, and the correlation between the first topic and another second topic B is 37.5%; if the preset correlation threshold is 50%, there is a second topic A whose correlation with the first topic meets the requirement; if the preset correlation threshold is 80%, there is no second topic whose correlation with the first topic meets the requirement.

[0061] Step 3.3), if it exists, the node corresponding to the second topic whose relevance to the first topic meets the preset relevance threshold is determined as the target node from the node tree, and the node tree is updated by making the node corresponding to the first topic a child node of the target node.

[0062] If the first topic corresponds to node 1.1.1, and there is only one second topic whose relevance to the first topic meets the preset relevance threshold, and its corresponding target node is node 1.1, then node 1.1.1 corresponding to the first topic is used as a child node of the target node, and the node tree is updated, such as Figure 2 As shown; if the first topic corresponds to node 1.1.2, and there are multiple second topics whose relevance to the first topic meets the preset relevance threshold, and their corresponding target nodes include nodes 1.1 and 1.2, then the node 1.1.2 corresponding to the first topic is used as a child node of the target node, and the node tree is updated, such as Figure 2 shown.

[0063] Step 3.4), if it does not exist, update the node tree by taking the node corresponding to the first topic as the root node.

[0064] If the first topic corresponds to node 2, and there is no second topic whose relevance to the first topic meets the preset relevance threshold, then node 2 is used as the root node to form a new node tree, such as Figure 2 shown.

[0065] It can be understood that, in practical applications, as the node tree is updated, it may include at least one.

[0066] As an optional embodiment, before step S106, the second subject may be redefined to ensure that the update accuracy of the node tree can be determined more accurately; specifically, the method further includes: Step 4.1), based on the historical request information and the historical responses corresponding to the historical request information, jointly determine the second topic corresponding to each historical request information.

[0067] In some cases, a new question may be asked in response to a previous answer. Here, the embodiment of the present invention can also jointly determine a second topic with the historical request information and the historical answer, and then match the second topic with the first topic for relevance, and execute steps 3.1) to 3.4) of the aforementioned embodiment to further improve the update accuracy of the node tree.

[0068] As an optional embodiment, step S108 may determine the target historical request information according to the association between nodes in the updated node tree, specifically including: Step 5.1), according to the updated node tree, determine the new node of the node tree.

[0069] The new node may be understood as a new node corresponding to the current request information or the first topic, that is, the node that is not present in the node tree before the update.

[0070] In step 5.2), based on the newly generated node, target historical request information whose relevance to the current request information meets the preset requirements is determined.

[0071] Here, according to the new node, at least one parent node of the new node can be obtained from the updated node tree, and the historical request information corresponding to each parent node is the target historical request information.

[0072] As another optional embodiment, step S108 may also determine the target historical request information in another manner, specifically including: Step 6.1), according to the updated node tree, determine the new node of the node tree.

[0073] The new node may be understood as a new node corresponding to the current request information or the first topic, that is, the node that is not present in the node tree before the update.

[0074] Step 6.2), based on the newly generated node and the historical request information corresponding to each node with which the newly generated node has a dependency relationship, determine the target historical request information whose correlation with the current request information meets the preset requirements.

[0075] Here, all the historical request information corresponding to the nodes directly connected to the new node or the upper nodes with dependency relationship can be taken as the historical request information. The dependency relationship here can be understood as all the paths starting from the root node and ending at the new node.

[0076] like Figure 2 As shown, if the new node is node 1.1.2, then based on the updated node tree, the historical request information corresponding to nodes 1.1, 1.2, and 1 that have a dependency relationship with node 1.1.2 can all be used as target historical request information whose correlation with the current request information meets the preset requirements.

[0077] In some embodiments, step S110 may be implemented by the following steps, including: Step 7.1), semantically identify the current summary information, the target historical request information, the historical summary information of the target real-time news event corresponding to the target historical request information, and the historical response, and determine the response to the current request information under the action of the multimodal large model.

[0078] The multimodal large model can fuse the summary, response and other information corresponding to the target historical requests that are associated and dependent with the current request information, as well as the summary information of the target real-time news events corresponding to the current request information, to parse out the accurate response to the current request information.

[0079] In some embodiments, Figure 3 As shown, the embodiment of the present invention provides a multi-topic dialogue processing device for real-time news based on a large model, comprising: A first determination module, in response to current request information inputted into the graphical user interface, determines a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event; a second determination module, which determines, from the real-time news events, a target real-time news event corresponding to the current request information and current summary information corresponding to the target real-time news event, according to the authenticity of each real-time news event and each first summary information; A matching module is used to match the first topic with the second topic corresponding to each historical request information, and update the node tree according to the matching result; wherein the node tree includes at least one node, the node is used to represent the topic corresponding to each request information, and the node tree is used to represent the relevance of each request information for real-time news events; A third determination module determines, based on the updated node tree, target historical request information whose relevance to the first topic meets preset requirements; The response module inputs the current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information into the multimodal large model, and outputs a response to the current request information.

[0080] Furthermore, the matching module is also specifically used to match the first topic with the second topic corresponding to each of the historical request information according to the neural network; determine whether there is a second topic whose correlation with the first topic meets the preset correlation threshold according to the matching result; if so, determine the node corresponding to the second topic whose correlation with the first topic meets the preset correlation threshold from the node tree as the target node, and update the node tree by using the node corresponding to the first topic as the child node of the target node; if not, update the node tree by using the node corresponding to the first topic as the root node.

[0081] Furthermore, the matching module is also specifically used to perform word segmentation processing on the first topic and the second topic corresponding to each historical request information through a neural network, and determine the word segmentation types of the first topic and the second topic; wherein the word segmentation types include entities and connecting relationship words; based on the word segmentation types, the first topic and the second topic are respectively matched in terms of relevance.

[0082] Furthermore, the matching module is specifically used to jointly determine the second topic corresponding to each historical request information based on the historical request information and the historical response corresponding to the historical request information before performing correlation matching between the first topic and the second topic corresponding to each historical request information and updating the node tree according to the matching result.

[0083] Furthermore, the third determination module is also specifically used to determine a new node of the node tree according to the updated node tree; based on the new node, determine the target historical request information whose correlation with the current request information meets the preset requirements.

[0084] Furthermore, the third determination module is also specifically used to determine the new node of the node tree according to the updated node tree; based on the new node and the historical request information corresponding to each node with which the new node has a dependency relationship, determine the target historical request information whose correlation with the current request information meets the preset requirements.

[0085] Furthermore, the first determination module is also specifically used to, in response to current request information input into the graphical user interface, control the real-time search engine to retrieve at least one real-time news event corresponding to the current request information; wherein the current request information includes a first request statement for locating at least one real-time news event and a second request statement for characterizing a corresponding request for the real-time news event; the first request statement and the second request statement are processed to determine a first topic corresponding to the current request information; wherein the first topic is composed of entities and connecting relationship words; corresponding key information is extracted from each of the real-time news events, and the key information is input into a multimodal big model to determine a first summary information corresponding to each of the real-time news events; wherein the multimodal big model includes a first preset template for extracting the first summary information, and the first preset template is determined according to the format requirements and attribute requirements of the first summary information.

[0086] Furthermore, the second determination module is also specifically used to determine the authenticity of the real-time news event corresponding to each of the first summary information according to the verification rules and verification weights corresponding to each news element of the first summary information in the second preset template of the multimodal large model; if the authenticity meets the requirements, the real-time news event is determined as a news fact, and according to the qualifiers in the first summary information, the target real-time news event corresponding to the current request information is determined from the news facts, and the first summary information corresponding to the target real-time news event is used as the current summary information; if the authenticity does not meet the requirements, the real-time news event is determined as a news clue, and the news clue in the real-time news event is removed.

[0087] Furthermore, the response module is also specifically used to perform semantic recognition on the current summary information, the target historical request information, the historical summary information and historical responses of the target real-time news event corresponding to the target historical request information, and determine the response to the current request information under the action of the multimodal large model.

[0088] 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.

[0089] 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.

[0090] 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.

[0091] 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.

[0092] 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 multi-topic dialogue processing method for real-time news based on a large model, It is characterized in that include: In response to current request information inputted into the graphical user interface, determining a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event; According to the authenticity of each of the real-time news events and each of the first summary information, determining a target real-time news event corresponding to the current request information and current summary information corresponding to the target real-time news event from the real-time news events; Match the first topic with the second topic corresponding to each historical request information, and update the node tree according to the matching result; wherein the node tree includes at least one node, the node is used to represent the topic corresponding to each request information, and the node tree is used to represent the relevance of each request information for real-time news events; Determine, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements; The current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information are input into a multimodal large model, and a response to the current request information is output.

2. The method according to claim 1, It is characterized in that The step of matching the first topic with the second topic corresponding to each historical request information by relevance, and updating the node tree according to the matching result, comprises: Matching the first topic with the second topic corresponding to each of the historical request information according to the neural network; Determining, based on the matching result, whether there is a second topic whose relevance to the first topic meets a preset relevance threshold; If so, determine the node corresponding to the second topic whose relevance to the first topic meets the preset relevance threshold from the node tree as the target node, and update the node tree by taking the node corresponding to the first topic as the child node of the target node; If it does not exist, the node tree is updated by taking the node corresponding to the first theme as the root node.

3. The method according to claim 2, It is characterized in that The step of matching the first topic with the second topic corresponding to each of the historical request information according to the neural network includes: Performing word segmentation processing on the first topic and the second topic corresponding to each historical request information respectively through a neural network to determine the word segmentation types of the first topic and the second topic; wherein the word segmentation types include entities and connecting relationship words; Based on the word segmentation type, the first topic and the second topic are respectively matched for relevance.

4. The method according to claim 1, It is characterized in that Before the step of matching the first subject with the second subject corresponding to each piece of historical request information and updating the node tree according to the matching result, the method further includes: A second topic corresponding to each piece of historical request information is jointly determined according to the historical request information and the historical responses corresponding to the historical request information.

5. The method according to claim 1, It is characterized in that The step of determining, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements comprises: According to the updated node tree, determine the new node of the node tree; Based on the new node, target historical request information whose relevance to the current request information meets preset requirements is determined.

6. The method according to claim 1, It is characterized in that The step of determining, according to the updated node tree, target historical request information whose relevance to the first topic meets preset requirements comprises: According to the updated node tree, determine the new node of the node tree; Based on the new node and the historical request information corresponding to each node having a dependent relationship with the new node, target historical request information whose correlation with the current request information meets preset requirements is determined.

7. The method according to claim 1, It is characterized in that In response to current request information inputted into a graphical user interface, the step of determining a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event comprises: In response to current request information inputted into the graphical user interface, controlling the real-time search engine to retrieve at least one real-time news event corresponding to the current request information; wherein the current request information includes a first request statement for locating at least one real-time news event and a second request statement for representing a corresponding request for the real-time news event; Processing the first request statement and the second request statement to determine a first topic corresponding to the current request information; wherein the first topic is composed of an entity and a connecting relation word; Corresponding key information is extracted from each of the real-time news events, and the key information is input into a multimodal big model to determine the first summary information corresponding to each of the real-time news events; wherein the multimodal big model includes a first preset template for extracting the first summary information, and the first preset template is determined according to the format requirements and attribute requirements of the first summary information.

8. The method according to claim 1, It is characterized in that The step of determining a target real-time news event corresponding to the current request information and the current summary information corresponding to the target real-time news event from the real-time news events according to the authenticity of each real-time news event and each first summary information comprises: Determine the authenticity of the real-time news event corresponding to each of the first summary information according to the verification rules and verification weights corresponding to each news element of the first summary information in the second preset template of the multimodal large model; If the authenticity meets the requirements, the real-time news event is determined as a news fact, and according to the qualifier in the first summary information, a target real-time news event corresponding to the current request information is determined from the news facts, and the first summary information corresponding to the target real-time news event is used as the current summary information; If the authenticity does not meet the requirements, the real-time news event is determined as a news clue, and the news clue in the real-time news event is removed.

9. The method according to claim 1, It is characterized in that The step of inputting the current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information into a multimodal macro model, and outputting a response to the current request information comprises: The current summary information, the target historical request information, the historical summary information and the historical response of the target real-time news event corresponding to the target historical request information are semantically identified, and the response to the current request information is determined under the action of the multimodal large model.

10. A multi-topic dialogue processing device for real-time news based on a large model, It is characterized in that include: A first determination module, in response to current request information inputted into the graphical user interface, determines a first topic corresponding to the current request information, at least one real-time news event, and first summary information corresponding to each real-time news event; a second determination module, which determines, from the real-time news events, a target real-time news event corresponding to the current request information and current summary information corresponding to the target real-time news event, according to the authenticity of each real-time news event and each first summary information; A matching module is used to match the first topic with the second topic corresponding to each historical request information, and update the node tree according to the matching result; wherein the node tree includes at least one node, the node is used to represent the topic corresponding to each request information, and the node tree is used to represent the relevance of each request information for real-time news events; A third determination module determines, based on the updated node tree, target historical request information whose relevance to the first topic meets preset requirements; The response module inputs the current request information, the current summary information, the target historical request information, and the historical summary information and historical responses corresponding to the target historical request information into the multimodal large model, and outputs a response to the current request information.

Images