Information processing method and device, equipment, storage medium and product

Abstract

The invention relates to an information processing method and device, equipment, a storage medium and a product, and the information processing method comprises the steps: receiving input multi-source knowledge information which comprises knowledge information from at least two information sources; under the condition that the multi-source knowledge information contains a conflict information group, based on an information field to which the knowledge information in the conflict information group belongs, performing conflict resolution processing on the knowledge information in the conflict information group to obtain target knowledge information; different information fields correspond to different conflict resolution modes; the conflict information group comprises a plurality of pieces of knowledge information with different information sources and information conflicts; and outputting the target knowledge information for the large model to call and perform question and answer reply. Through the method, the limitation of a traditional single rule in processing multi-source heterogeneous knowledge is effectively overcome, and the processing accuracy of knowledge conflicts in different fields is remarkably improved.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to an information processing method, apparatus, device, storage medium and product. Background Technology

[0002] With the development of artificial intelligence technology, knowledge bases have become a crucial foundation for large-scale model applications. However, current knowledge information exhibits typical characteristics of being multi-source, heterogeneous, and dynamically evolving. The same entity often has differentiated or even conflicting descriptions from different sources such as guidelines, literature, and practices. To ensure the accuracy and reliability of the output content of large-scale models, it is essential to conduct rigorous credibility assessments and conflict resolution on the knowledge incorporated into model training and inference.

[0003] Existing knowledge processing solutions mainly rely on using a single, fixed priority rule to resolve knowledge conflicts. This approach cannot adapt to the changing needs of dynamic scenarios, resulting in low accuracy of large models when responding to questions based on knowledge information. Summary of the Invention

[0004] This application provides an information processing method, apparatus, device, storage medium, and product, which effectively overcomes the limitations of traditional single rules in processing multi-source heterogeneous knowledge and significantly improves the accuracy of handling knowledge conflicts in different domains. The technical solution is as follows.

[0005] On the one hand, an information processing method is provided, the method comprising: Receive input multi-source knowledge information, wherein the multi-source knowledge information includes knowledge information from at least two information sources; When the multi-source knowledge information includes conflicting information groups, the knowledge information in the conflicting information groups is processed to resolve conflicts based on the information domain to which the knowledge information in the conflicting information groups belongs, thereby obtaining the target knowledge information; different information domains correspond to different conflict resolution methods; the conflicting information groups contain multiple knowledge information groups with different information sources and conflicting information. The target knowledge information is output for use by the large model to perform question-and-answer responses.

[0006] On the other hand, an information processing apparatus is provided, the apparatus comprising: The information receiving module is used to receive input multi-source knowledge information, wherein the multi-source knowledge information includes knowledge information from at least two information sources; The conflict resolution module is used to resolve conflicts in the multi-source knowledge information group based on the information domain to which the knowledge information in the conflict information group belongs, when such conflict information group is contained in the multi-source knowledge information group, in order to obtain the target knowledge information; different information domains correspond to different conflict resolution methods; the conflict information group contains multiple knowledge information groups with different information sources and conflicting information; The first information output module is used to output the target knowledge information for the large model to call and respond to questions.

[0007] In one possible implementation, the conflict resolution method includes: timeliness priority, authority priority, or majority consensus assessment; the conflict resolution module is used for, If the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is time-sensitive, then the knowledge information in the conflict information group that is closest to the current time is determined as the target knowledge information. When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is authority priority, the knowledge information with the highest authority score among the information sources in the conflict information group is determined as the target knowledge information; the authority score is a score obtained by analyzing the reference network between information sources based on an authority score algorithm; When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is majority consensus assessment, the knowledge information in the conflict information group that appears more frequently than a frequency threshold is identified as the target knowledge information.

[0008] In one possible implementation, the device further includes: The second information output module is used to output conflict information when the multi-source knowledge information contains conflict information groups. The conflict information includes at least one of the following: conflict type, conflict knowledge, and conflict level. The conflict type includes numerical conflict or logical conflict. The conflict level is determined based on preset conflict level rules, and the conflict level rules are different for different information domains.

[0009] In one possible implementation, the device further includes: The preprocessing module is used to preprocess the multi-source knowledge information after the information receiving module receives the input multi-source knowledge information to obtain standardized knowledge information corresponding to each knowledge information. The credibility assessment module is used to conduct multi-dimensional credibility assessments of standardized knowledge information for various knowledge information, and obtain credibility assessment results for each dimension. The comprehensive credibility determination module is used to comprehensively determine the comprehensive credibility of the corresponding knowledge information based on the credibility evaluation results of each knowledge information in multiple dimensions. The information filtering module is used to filter information based on the overall credibility of various knowledge information to obtain filtered multi-source knowledge information.

[0010] In one possible implementation, the multi-dimensional credibility assessment includes at least two of the following: information source authority assessment, timeliness decay assessment, and knowledge consistency assessment. The credibility assessment module is used for, When the multi-dimensional credibility assessment includes an assessment of the authority of the information source, an authority score for the knowledge information is output based on the information source identifier in the standardized knowledge information and a pre-constructed authority graph. When the multi-dimensional credibility assessment includes a timeliness decay assessment, the timeliness score of the knowledge information is calculated based on the information release time and the decay factor of the information domain to which the knowledge information belongs in the standardized knowledge information. In the case where the multi-dimensional credibility assessment includes knowledge consistency assessment, a consistency score between knowledge information and known knowledge information is calculated using a graph attention network based on a pre-constructed knowledge graph, wherein the knowledge graph contains entity nodes, relation edges, and time information.

[0011] In one possible implementation, the comprehensive credibility determination module is used to perform a weighted summation of the credibility evaluation results of multiple dimensions to obtain the comprehensive credibility of the knowledge information; the weighting weights corresponding to the credibility evaluation of different dimensions are different.

[0012] In one possible implementation, the device further includes: The adjustment module is used to adjust the weighting of the multi-dimensional credibility assessment based on the user feedback information received based on the target knowledge information.

[0013] In one possible implementation, the device further includes: The information addition module is used to add non-conflicting knowledge information from the multi-source knowledge information to the knowledge base for large models to call and respond to questions.

[0014] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to implement the above-described information processing method.

[0015] On the other hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the above-described information processing method.

[0016] On the other hand, a computer program product is provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform to implement the information processing method provided in the various optional implementations described above.

[0017] The information processing method provided in this application receives multi-source knowledge information from different information sources. When conflicting information groups are present within the multi-source knowledge information, the method resolves the conflicts in the conflicting information groups based on a conflict resolution method corresponding to the information domain to which the conflicting information groups belong. This results in conflict-resolved target knowledge information, which is then used by a large model for question-and-answer responses. By introducing an information domain-adaptive conflict resolution mechanism, the limitations of traditional single rules in processing multi-source heterogeneous knowledge are effectively overcome, significantly improving the accuracy of handling knowledge conflicts across different domains. The conflict-resolved knowledge information provides a high-quality, conflict-free training and reasoning foundation for the large model, thereby improving the overall professional accuracy and user credibility of the generated answers.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] Figure 1 A flowchart of an exemplary embodiment of the present application is shown; Figure 2 A flowchart of an information processing method provided in another embodiment of this application is shown; Figure 3 This invention provides a structural block diagram of an information processing apparatus according to an exemplary embodiment of the present application. Figure 4 This application shows a structural block diagram of a computer device according to an exemplary embodiment. Figure 5 A structural block diagram of a computer device is shown in another exemplary embodiment of this application. Detailed Implementation

[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of methods consistent with some aspects of this application as detailed in the appended claims.

[0022] Large-scale models refer to basic artificial intelligence models with a large number of parameters built by training on massive and diverse general datasets, possessing versatility. Based on large-scale models, training and optimization for specific data and tasks can yield large-scale models with specialized knowledge and capabilities, i.e., industry-specific large-scale models. These industry-specific large-scale models can be applied to corresponding industry domains, improving their application effectiveness. Whether it's a large-scale model or an industry-specific large-scale model, the accuracy of its question-and-answer responses depends on the retrieved knowledge information. However, due to the diversity of knowledge information sources, knowledge conflicts may exist between different sources. To improve the effectiveness of resolving knowledge conflicts, adapt to dynamic changes in scenario requirements, and enhance the accuracy of large-scale models in question-and-answer responses based on knowledge information, this application provides an information processing method. Figure 1 This application illustrates a flowchart of an information processing method provided in an exemplary embodiment. This method can be executed by a computer device, which can be implemented as a server or a terminal, such as... Figure 1 As shown, the information processing method may include the following steps.

[0023] Step 110: Receive input multi-source knowledge information, which contains knowledge information from at least two information sources.

[0024] Knowledge information from different sources may have different representational structures. Schematic, multi-source knowledge information can include structured data, semi-structured data, and unstructured data. Structured data refers to data with a high degree of organization and a fixed format, usually stored in database tables, and has high readability. Unstructured data refers to data without a predefined data model or organizational form, such as guides, papers, user-generated content, images, videos, etc. Semi-structured data is between structured and unstructured data. Semi-structured data can carry metadata describing its content to separate data elements, giving it a certain hierarchical structure. For example, metadata can be tags, fields, etc.

[0025] In one possible implementation, since the representational structures of knowledge information from different sources may differ, to facilitate subsequent information processing, the computer device, after receiving input multi-source knowledge information, can preprocess the multi-source knowledge information to generate structured information for each piece of knowledge. Then, subsequent processing is performed based on this structured information. Furthermore, before conflict verification and resolution, the computer device can also perform a credibility assessment on the received knowledge information to filter it, retaining high-credibility knowledge information for subsequent operations. This reduces the amount of data to be processed while preserving the high credibility of the knowledge information. The above process can be implemented as follows: Preprocessing of multi-source knowledge information yields structured information groups for each piece of knowledge information; A multi-dimensional credibility assessment is conducted on the structured information groups of various knowledge information to obtain the credibility assessment results of each dimension; The overall credibility of the corresponding knowledge information is determined by comprehensively evaluating the credibility of each knowledge information across multiple dimensions. Information is filtered based on the overall credibility of various knowledge sources to obtain filtered multi-source knowledge information.

[0026] Preprocessing multi-source knowledge information can include information standardization. Information standardization refers to identifying the source, timestamp, and other information of the knowledge information after receiving the knowledge data to obtain standardized knowledge information. This standardized knowledge information may include <knowledge information content, information source identifier, and timestamp> extracted according to a standard format. The information source identifier is used to indicate the source of the knowledge information. Furthermore, based on different actual needs, the quantity and type of standardized knowledge information can be adjusted. For example, the standardized knowledge information may also include the original confidence level of the knowledge information, author institution, knowledge type, etc. The original confidence level is determined based on the source of the knowledge information. Generally, the higher the authority of the source of the knowledge information, the higher its original confidence level. This application embodiment does not limit the information content contained in the standardized knowledge information obtained after information standardization processing. By performing information standardization processing on the received knowledge information, subsequent knowledge information processing and information extraction can be facilitated.

[0027] Furthermore, after the information is standardized, the computer device can perform structured information extraction based on the standardized knowledge information. This structured information extraction can include entity extraction, entity relationship identification, and information validity time extraction, thereby obtaining a structured information group of the corresponding knowledge information. This structured information group can be illustrated as a triple of <head entity, relationship, tail entity, time interval>. It should be noted that a knowledge information can contain one or more entities. The above structured information group is only illustrative, and this application does not limit the number of entities in the structured information group. In addition, the same knowledge information can generate one or more structured information groups. Taking a drug instruction manual as an example, if the entities obtained after structured information extraction are drug A, disease B, and drug C, the entity relationship between drug A and disease B is "treatment", the entity relationship between drug A and drug C is "interaction", and the information validity time is effective in 2023, then the structured information group 1 of the drug instruction manual can be obtained as <drug A, treatment, disease B, effective on: 2023-05-10>, and the structured information group 2 as <drug A, interaction, drug C>. Effective from: 2023-05-10> By structuring various knowledge information, it is easier to extract and query information later.

[0028] This application's embodiments can assess the credibility of knowledge information from multiple dimensions and comprehensively determine the credibility of the knowledge information based on the credibility assessment results of each dimension. In one possible implementation, the multi-dimensional credibility assessment includes at least two of the following: information source authority assessment, timeliness decay assessment, and knowledge consistency assessment. Specifically, the higher the authority of the information source, the higher the authority score of the corresponding knowledge information; the shorter the time since the information was published, the higher its timeliness score; and the higher the consistency with known information, the higher its corresponding consistency score. Furthermore, based on different actual needs, the credibility assessment can be extended to other dimensions, such as the publisher's reputation, the number of citations, etc.

[0029] A multi-dimensional credibility assessment is conducted on standardized knowledge information for each category, yielding credibility assessment results for each dimension, including: In the context of multi-dimensional credibility assessment, including assessment of the authority of information sources, an authority score for the knowledge information is output based on the information sources in standardized knowledge information and a pre-constructed authority graph. In the context of multi-dimensional credibility assessment, which includes timeliness decay assessment, the timeliness score of knowledge information is calculated based on the information release time and decay factor of the information domain to which the knowledge information belongs in the standardized knowledge information. In the context of multi-dimensional credibility assessment, which includes knowledge consistency assessment, a consistency score between knowledge information and known knowledge information is calculated using a graph attention network based on a pre-built knowledge graph. The knowledge graph contains entity nodes, relation edges, and time information.

[0030] The authority graph is a knowledge graph used to describe the authority relationships between various information sources. Each node in the authority graph represents an information source node, such as an article, institution, website, expert, database, etc. Edges in the authority graph can represent affiliation or publication relationships; for example, an expert may belong to a university, or an article may be published in a journal. Each information source node has a corresponding authority rating attribute, which can be assigned by professionals or calculated using the PageRank algorithm. The PageRank algorithm calculates authority ratings through a voting model, meaning that a node's authority depends on the quality and quantity of other nodes linked to it; the more high-authority nodes linking a node, the higher its authority rating. In this embodiment, when evaluating information source authority, the computer device can locate nodes in the authority graph based on the information source indication in the standardized information, and determine the authority rating of the located information source node as the authority rating of the knowledge information. It should be noted that this authority rating can be a normalized numerical value, i.e., the authority rating R∈[0,1].

[0031] In assessing the timeliness decay, computer equipment can calculate the decay duration based on the information's publication time. For example, if the information publication time is represented as t... publish The decay time Δt=t now -t publish , t now The decay factor of the information domain to which the knowledge information belongs, representing the current time, is denoted as λ. The formula for calculating the timeliness score β can be expressed as: β=exp(-λ×Δt) Among them, the timeliness score β∈(0,1] should be noted that the decay factor is different for different information domains. The decay factor for each information domain can be set according to the information update speed of each information domain. The faster the information update speed, the larger the corresponding decay factor value, and the slower the information update speed, the smaller the corresponding decay factor value. For example, the knowledge information in the medical field is updated relatively quickly, and its decay factor can be 0.15. The policy information in the financial field is updated in a moderate way, and its decay factor can be 0.08, etc. As the field develops and changes, the decay factor of the same information domain can also be dynamically adjusted to adapt to the changes in the knowledge update rate.

[0032] In the context of knowledge consistency assessment, computer devices can analyze the topological relationships of knowledge information in a knowledge graph to verify whether the knowledge information is compatible with existing knowledge information. This knowledge graph is a structured network constructed from known knowledge information from different sources, containing a large number of known entities, relationships, and their attributes. When calculating the consistency score between the knowledge information and known knowledge information using a graph attention network, the computer device can temporarily insert the knowledge information into the corresponding position in the knowledge graph based on the structured information. The graph attention network scans the local connectivity environment of the head and tail entity nodes in this structured information within the knowledge graph, analyzing the complex network structure formed by their relationships with surrounding entities. Through the attention mechanism, the network can adaptively learn the importance weights of different neighboring nodes for consistency judgments, and finally calculate a quantitative consistency score by integrating these weighted information, thereby objectively reflecting whether new knowledge corroborates or conflicts with the existing knowledge system. In other words, the graph attention network analyzes the correlation between each entity node of the knowledge information and each known connection relationship in the knowledge graph and the current connection relationship, and assigns corresponding weight values ​​based on the correlation between each connection relationship and the current connection relationship. The higher the correlation, the higher the corresponding weight value, and the lower the correlation, the lower the corresponding weight value. If the current connection relationship contradicts another connection relationship, its corresponding weight value can be negative. After determining the weight values ​​based on the correlation between each connection relationship and the current connection relationship, the computer device can perform comprehensive weighting and normalization to obtain the consistency score of the knowledge information. It should be noted that the computer device may have a preset mapping relationship between correlation and weight values. This mapping relationship can be a relevant calculation formula or an assignment rule. This application embodiment does not limit this.For example, if the structured information of knowledge is (drug A, causing, disease B), in the knowledge graph, for the node "drug A", its neighbor environment includes relation 1: (drug A -- treatment --> disease B), relation 2: (drug A -- inhibition --> protein X), and for the node "drug B", its neighbor environment includes relation 1: (drug A -- treatment --> disease B), relation 3: (drug C... --leading to-->disease B), since relation 1 directly contradicts the structured information of the knowledge information, its corresponding weight value may be a large negative value, such as -0.8. Relation 3 corresponds to the structured information of the knowledge information as similar information, and its corresponding weight value may be a moderate positive value, such as 0.4. Relation 2 has a low correlation with the structured information of the knowledge information, so it can be assigned a low weight value, such as 0.1. After obtaining the weights corresponding to each relation, the consistency score of the knowledge information can be obtained by weighted summation and normalization. In one possible implementation, the computer device can normalize using the Sigmoid function. Since the strength of the negative weight is much greater than that of the positive weight in the above example, a consistency score close to 0, such as 0.05, can be obtained after the Sigmoid function.

[0033] The overall credibility of the corresponding knowledge information is determined by comprehensively evaluating the credibility of each piece of knowledge information across multiple dimensions, including: The overall credibility of knowledge information is obtained by weighted summation of the credibility assessment results from multiple dimensions; the weighting weights are different for different dimensions of credibility assessment.

[0034] An illustrative representation of the calculation process for overall credibility can be expressed as follows: W k =aR+bβ+cγ,W k ∈[0,3] Here, a, b, and c are the weighted weights of each dimension, and the weighted weight values ​​of each dimension can be set based on actual needs. This application does not impose any restrictions on this.

[0035] Alternatively, in another possible implementation, the overall credibility can be the average of the credibility assessment results of multiple dimensions, or other feasible statistical values, which are not limited in this application embodiment.

[0036] After obtaining the overall credibility of various knowledge information, the computer device can filter information based on the overall credibility of each knowledge information to obtain filtered multi-source knowledge information. By filtering information based on overall credibility, the problem of poor response results caused by using low-credibility information as the basis for large model question and answer can be avoided, thereby improving the accuracy of large model question and answer.

[0037] After obtaining the filtered multi-source knowledge information, the computer equipment can also perform conflict verification on each filtered knowledge information, so as to further process the knowledge information based on the conflict verification results.

[0038] Step 120: In the case where there are conflicting information groups in the multi-source knowledge information, the conflict information groups are processed to resolve conflicts based on the information domain to which the knowledge information in the conflicting information groups belong, so as to obtain the target knowledge information; the conflict resolution methods are different for different information domains; the conflicting information group contains multiple knowledge information groups with different information sources and conflicting information.

[0039] The types of knowledge information conflicts can include numerical conflicts or logical conflicts. Numerical conflicts refer to discrepancies in the attribute values ​​of a unified entity; for example, knowledge information 1 indicates that the dosage of drug A is 10mg, while knowledge information 2 indicates that the dosage of drug A is 20mg. Logical conflicts refer to contradictory relationships expressed by knowledge information; for example, knowledge information 3 indicates that drug B is recommended to be taken after meals, while knowledge information 3 indicates that drug B is prohibited from being taken after meals. When two or more pieces of knowledge information conflict, the conflicting knowledge information is added to a conflict information group, and conflict resolution is performed on a group-by-group basis. Each conflict information group corresponds to a target knowledge information after conflict resolution.

[0040] In one possible implementation, when a computer device determines that multi-source knowledge information contains conflicting knowledge information, it can also extract and output the conflict information so that relevant personnel can view and manually correct the conflict information. This process can be implemented as follows: In the case of conflicting information groups contained in multi-source knowledge information, the conflicting information is output. The conflicting information includes at least one of the following: conflict type, conflicting knowledge, and conflict level. The conflict type includes numerical conflict or logical conflict, and the conflict level is determined based on preset conflict level rules. The conflict level rules are different for different information domains.

[0041] The classification of conflict levels can be based on the severity of the events caused by the conflict; the more severe the event, the higher the corresponding conflict level. Conflict level rules can be set based on actual needs, and this application does not impose any restrictions on this. These conflict level rules may include quantitative rules for conflict severity, the number of conflict levels, and the correspondence between conflict severity and conflict levels. Different information domains have different conflict level determination standards, meaning that the conflict level rules differ across information domains. For example, in the medical field, conflict levels can include three levels: Level L1 is a conflict involving life safety, such as a conflict over surgical contraindications; Level L2 is a conflict over general clinical recommendations, such as a conflict over medication frequency; and Level L3 is a conflict over non-critical data, such as a conflict over drug packaging specifications. In the above example, the smaller the numerical value corresponding to the conflict level, the higher the conflict level.

[0042] In a multi-source knowledge information system containing conflict information groups, and after identifying the conflict information, the conflict information is output. In one possible implementation, the computer device can output the conflict information in the form of a conflict report, which can be <conflict type, involved entities, conflict level>. The computer device can then feed back the conflict report to relevant personnel so that they can confirm or correct the conflict information.

[0043] After identifying conflicting information, a conflict resolution process is triggered. One possible implementation includes conflict resolution methods such as prioritizing timeliness, prioritizing authority, or majority consensus assessment. Based on the information domain to which the knowledge information in the conflicting information group belongs, conflict resolution processing is performed on the knowledge information in the conflicting information group to obtain the target knowledge information, including: When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is time-sensitive, the knowledge information in the conflict information group that is closest to the current time is identified as the target knowledge information. When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is authority-first, the knowledge information with the highest authority score among the information sources in the conflict information group is determined as the target knowledge information; the authority score is obtained by analyzing the reference network between information sources based on the authority score algorithm; When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is majority consensus assessment, the knowledge information in the conflict information group that appears more frequently than the frequency threshold is identified as the target knowledge information.

[0044] In other words, when timeliness is prioritized in conflict resolution, the computer device identifies the knowledge information with the shortest publication time among the conflicting knowledge information as the target knowledge information. When authority is prioritized, the computer device can identify the knowledge information with the highest authority score from the knowledge source to which the conflicting knowledge information belongs as the target knowledge information. When majority consensus assessment is used, the computer device can identify the most frequently occurring knowledge information in the conflicting information group, or multiple knowledge information with frequencies exceeding a frequency threshold, as the target knowledge information. If multiple knowledge information with frequencies exceeding the frequency threshold exist, in one possible implementation, the computer device can use all conflicting knowledge information as target knowledge information for the large model to perform question-and-answer responses. In this case, the large model may generate different responses based on the conflicting knowledge information. Subsequently, the frequency threshold can be adjusted based on user feedback on the responses to ensure that knowledge information receiving positive user feedback can be filtered out as target knowledge information. Optionally, the adjustment step of the frequency threshold can be ±5%. Alternatively, in another possible implementation, the computer device can send the conflicting knowledge information to relevant personnel for expert evaluation to determine the target knowledge information.

[0045] In one possible implementation, the computer device can resolve conflicts using a decision tree approach. Decision tree resolution involves sequentially determining the resolution method based on the relevant information domain, and then resolving the conflict according to the matching method. Illustratively, this process can be implemented as follows: Determine whether timeliness is a priority based on the information domain: if so, sort the knowledge information in reverse chronological order and select t. publish The latest version is the target knowledge information; if not, then an authority-priority evaluation will be conducted. The evaluation criteria are determined based on the information domain: if so, knowledge information from highly authoritative data sources is selected as the target knowledge information by comparing authority scores; if not, i.e., there is no critical conflict and no significant difference in authority, the evaluation proceeds to the majority consensus assessment. Most consensus assessments statistically analyze the number of sources supporting different values ​​and select knowledge information whose frequency exceeds a frequency threshold as the target knowledge information.

[0046] It should be noted that in the process of conflict resolution using decision tree resolution, the order of judgment of various conflict resolution methods can be adjusted based on actual needs, and this application embodiment does not impose any restrictions on this.

[0047] Step 130: Output the target knowledge information for the large model to call and respond to questions.

[0048] In one possible implementation, if the large model responds to questions based on the target knowledge information after conflict resolution and receives feedback from the user based on that response, the computer device can adjust the credibility assessment of the knowledge information based on the user feedback. This process can be implemented as follows: In response to receiving user feedback information based on target knowledge information, the weighted weights of the multi-dimensional credibility assessment are adjusted based on the user feedback information.

[0049] In other words, if the user feedback received is negative, it means that the credibility of the knowledge information on which the current response is based is low. Therefore, the computer device can adjust the multi-dimensional credibility weighting to reduce the overall credibility of the knowledge information on which the current response is based, thereby preventing it from being filtered out. Optionally, the adjustment range of the weighting can be [-0.1, 0.1].

[0050] Optionally, the computer device can be configured with a multi-objective reward function to evaluate the screening effect and guide the learning direction. This reward function can consist of user satisfaction rewards, post-validation rewards, and action magnitude penalties. The reward function can be expressed as:

[0051] Among them, user satisfaction rewards This refers to normalizing user feedback to the interval [-1, 1]. If the user feedback is a feedback rating with a value range of 1-5, then the user satisfaction reward can be represented as r. user = (User rating - 3) / 2; Post-verification reward This refers to awarding a positive reward when the target knowledge information output by the system matches the target knowledge information determined by domain experts, which can be represented as r. valid = I(System Recommendation == Expert Ruling), I() is the indicator function, and the action amplitude penalty is ||a t ||2 is the L2 norm of the action vector, used to constrain the magnitude of the policy output and prevent over-optimization of a single objective; based on the above reward function, it guides the adjustment of the confidence formula weights, or indicates the modification of the decision tree threshold adjustment, so that the knowledge information processing is more in line with the requirements.

[0052] In one possible implementation, the computer device can also add non-conflicting knowledge information from multiple sources to the knowledge base for large models to call and respond to questions.

[0053] In other words, non-conflicting knowledge information from multi-source knowledge information that has undergone comprehensive credibility screening is added to the knowledge base to serve as the basis for the large model to respond to questions.

[0054] In summary, the information processing method provided in this application receives multi-source knowledge information from different information sources. When conflicting information groups are present within these multi-source knowledge information, the method resolves the conflicts in the knowledge information groups based on a conflict resolution method corresponding to the information domain to which the conflicting information groups belong. This results in conflict-resolved target knowledge information, which is then used by a large model for question-and-answer responses. By introducing an information domain-adaptive conflict resolution mechanism, the limitations of traditional single-rule methods in processing multi-source heterogeneous knowledge are effectively overcome, significantly improving the accuracy of handling knowledge conflicts across different domains. The conflict-resolved knowledge information provides a high-quality, conflict-free training and reasoning foundation for the large model, thereby improving the overall professional accuracy and user credibility of the generated answers.

[0055] Taking computer devices using decision trees to resolve conflicts as an example, Figure 2 A flowchart of an information processing method provided in another embodiment of this application is shown, such as... Figure 2 As shown, the method may include the following steps.

[0056] S201 receives input multi-source knowledge information.

[0057] S202, preprocess the multi-source knowledge information to obtain standardized knowledge information for each knowledge source.

[0058] S203, conducts multi-dimensional credibility assessments on standardized knowledge information for each knowledge information, and obtains credibility assessment results for each dimension.

[0059] Multi-dimensional credibility assessment includes assessment of the authority of the information source, assessment of timeliness decay, and assessment of knowledge consistency.

[0060] S204. The overall credibility of the corresponding knowledge information is determined based on the credibility assessment results of each knowledge information in multiple dimensions.

[0061] S205, based on the comprehensive credibility of various knowledge information, information is filtered to obtain filtered multi-source knowledge information.

[0062] S206, Conflict detection is performed on multi-source knowledge information.

[0063] S207, If a conflict information group is included, initiate the decision tree.

[0064] S208. Determine whether timeliness takes precedence based on the information domain to which the conflicting information group belongs. If so, proceed to S209; otherwise, proceed to S210.

[0065] S209, Select the latest version as the target knowledge information.

[0066] S210: Determine whether the information domain to which the conflicting information group belongs is of priority based on authority. If so, execute S211; otherwise, execute S212.

[0067] S211, identify knowledge information from highly authoritative information sources as target knowledge information.

[0068] S212, conduct a majority consensus assessment to obtain the target knowledge information.

[0069] S213, output target knowledge information.

[0070] S214, Add non-conflicting knowledge information from multi-source knowledge information to the knowledge base.

[0071] S215 receives user feedback information.

[0072] S216 uses reinforcement learning to optimize based on user feedback and adjusts system parameters.

[0073] The system parameter adjustments include adjusting the weighting of multi-dimensional credibility assessments and adjusting the frequency thresholds in most consensus assessments.

[0074] Figure 2 For details regarding each step shown, please refer to the following: Figure 1 The details of the corresponding steps shown are not repeated here.

[0075] Figure 3 This application shows a structural block diagram of an information processing apparatus provided in an exemplary embodiment. This apparatus can be applied in a computer device for performing tasks such as... Figure 1 or Figure 2 All or part of the steps in the illustrated embodiments, such as Figure 3 As shown, the device may include the following modules.

[0076] Information receiving module 310 is used to receive input multi-source knowledge information, wherein the multi-source knowledge information includes knowledge information from at least two information sources; The conflict resolution module 320 is used to resolve conflicts in the knowledge information in the multi-source knowledge information group based on the information domain to which the knowledge information in the conflict information group belongs, in order to obtain the target knowledge information; different information domains correspond to different conflict resolution methods; the conflict information group contains multiple knowledge information from different information sources and which conflict with each other; The first information output module 330 is used to output the target knowledge information for the large model to call and perform question and answer responses.

[0077] In one possible implementation, the conflict resolution method includes: timeliness priority, authority priority, or majority consensus assessment; the conflict resolution module 320 is used for, If the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is time-sensitive, then the knowledge information in the conflict information group that is closest to the current time is determined as the target knowledge information. When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is authority priority, the knowledge information with the highest authority score among the information sources in the conflict information group is determined as the target knowledge information; the authority score is a score obtained by analyzing the reference network between information sources based on an authority score algorithm; When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is majority consensus assessment, the knowledge information in the conflict information group that appears more frequently than a frequency threshold is identified as the target knowledge information.

[0078] In one possible implementation, the device further includes: The second information output module is used to output conflict information when the multi-source knowledge information contains conflict information groups. The conflict information includes at least one of the following: conflict type, conflict knowledge, and conflict level. The conflict type includes numerical conflict or logical conflict. The conflict level is determined based on preset conflict level rules, and the conflict level rules are different for different information domains.

[0079] In one possible implementation, the device further includes: The preprocessing module is used to preprocess the multi-source knowledge information after the information receiving module receives the input multi-source knowledge information to obtain standardized knowledge information corresponding to each knowledge information. The credibility assessment module is used to conduct multi-dimensional credibility assessments of standardized knowledge information for various knowledge information, and obtain credibility assessment results for each dimension. The comprehensive credibility determination module is used to comprehensively determine the comprehensive credibility of the corresponding knowledge information based on the credibility evaluation results of each knowledge information in multiple dimensions. The information filtering module is used to filter information based on the overall credibility of various knowledge information to obtain filtered multi-source knowledge information.

[0080] In one possible implementation, the multi-dimensional credibility assessment includes at least two of the following: information source authority assessment, timeliness decay assessment, and knowledge consistency assessment. The credibility assessment module is used for, When the multi-dimensional credibility assessment includes an assessment of the authority of the information source, an authority score for the knowledge information is output based on the information source identifier in the standardized knowledge information and a pre-constructed authority graph. When the multi-dimensional credibility assessment includes a timeliness decay assessment, the timeliness score of the knowledge information is calculated based on the information release time and the decay factor of the information domain to which the knowledge information belongs in the standardized knowledge information. In the case where the multi-dimensional credibility assessment includes knowledge consistency assessment, a consistency score between knowledge information and known knowledge information is calculated using a graph attention network based on a pre-constructed knowledge graph, wherein the knowledge graph contains entity nodes, relation edges, and time information.

[0081] In one possible implementation, the comprehensive credibility determination module is used to perform a weighted summation of the credibility evaluation results of multiple dimensions to obtain the comprehensive credibility of the knowledge information; the weighting weights corresponding to the credibility evaluation of different dimensions are different.

[0082] In one possible implementation, the device further includes: The adjustment module is used to adjust the weighting of the multi-dimensional credibility assessment based on the user feedback information received based on the target knowledge information.

[0083] In one possible implementation, the device further includes: The information addition module is used to add non-conflicting knowledge information from the multi-source knowledge information to the knowledge base for large models to call and respond to questions.

[0084] In summary, the information processing apparatus provided in this application receives multi-source knowledge information from different information sources. When the multi-source knowledge information contains conflicting information groups, it resolves the conflicts in the knowledge information groups based on a conflict resolution method corresponding to the information domain to which the conflicting information groups belong, obtaining the conflict-resolved target knowledge information for use by a large model for question-and-answer responses. By introducing an information domain-adaptive conflict resolution mechanism, the limitations of traditional single rules in processing multi-source heterogeneous knowledge are effectively overcome, significantly improving the accuracy of handling knowledge conflicts across different domains. The conflict-resolved knowledge information provides a high-quality, conflict-free training and reasoning foundation for the large model, thereby improving the overall professional accuracy and user credibility of the generated answers.

[0085] Figure 4This diagram illustrates a structural block diagram of a computer device 400 according to an exemplary embodiment of this application. This computer device can be implemented as a server as described in the above-described scheme of this application. The computer device 400 includes a Central Processing Unit (CPU) 401, a system memory 404 including Random Access Memory (RAM) 402 and Read-Only Memory (ROM) 403, and a system bus 405 connecting the system memory 404 and the CPU 401. The computer device 400 also includes a mass storage device 406 for storing an operating system 409, application programs 410, and other program modules 411. The system memory 404 and the mass storage device 406 can be collectively referred to as memory.

[0086] According to various embodiments of this application, the computer device 400 can also be connected to a remote computer on a network, such as the Internet. That is, the computer device 400 can be connected to a network 408 via a network interface unit 407 connected to the system bus 405, or the network interface unit 407 can be used to connect to other types of networks or remote computer systems (not shown).

[0087] The memory also includes at least one instruction, at least one program, code set, or instruction set, which are stored in the memory. The central processing unit 401 executes the at least one instruction, at least one program, code set, or instruction set to implement all or part of the steps in the information processing methods shown in the above embodiments.

[0088] Figure 5 A structural block diagram of a computer device 500 illustrating another exemplary embodiment of this application is shown. The computer device 500 can be implemented as the aforementioned terminal. For example, the computer device can be an Android terminal device; typically, the computer device 500 includes a processor 501 and a memory 502. The memory 502 may include one or more computer-readable storage media for storing at least one instruction, which is executed by the processor 501 to implement all or part of the steps in the information processing method shown in the method embodiments of this application.

[0089] In some embodiments, the computer device 500 may optionally include a peripheral device interface 503 and at least one peripheral device. The processor 501, memory 502, and peripheral device interface 503 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 503 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 504, a display screen 505, a camera assembly 506, an audio circuit 507, and a power supply 508. In some embodiments, the computer device 500 further includes one or more sensors 509. These sensors 509 include, but are not limited to, an accelerometer 510, a gyroscope 511, a pressure sensor 512, an optical sensor 513, and a proximity sensor 514. Those skilled in the art will understand that... Figure 5 The structure shown does not constitute a limitation on the computer device 500, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0090] In one exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one computer program that is loaded and executed by a processor to implement all or part of the steps in the information processing method described above. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, or optical data storage device, etc.

[0091] In one exemplary embodiment, a computer program product is also provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, cause the computer to perform the above-described actions. Figure 1 or Figure 2 All or part of the steps of the embodiments shown in any of the embodiments.

[0092] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0093] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. An information processing method, characterized in that, The method includes: Receive input multi-source knowledge information, wherein the multi-source knowledge information includes knowledge information from at least two information sources; When the multi-source knowledge information includes conflicting information groups, the knowledge information in the conflicting information groups is processed to resolve conflicts based on the information domain to which the knowledge information in the conflicting information groups belongs, thereby obtaining the target knowledge information; different information domains correspond to different conflict resolution methods; the conflicting information groups contain multiple knowledge information groups with different information sources and conflicting information. The target knowledge information is output for use by the large model to perform question-and-answer responses.

2. The method according to claim 1, characterized in that, The conflict resolution methods include: timeliness priority, authority priority, or majority consensus assessment; the conflict resolution process, based on the information domain to which the knowledge information in the conflict information group belongs, performs conflict resolution processing on the knowledge information in the conflict information group to obtain the target knowledge information, including: If the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is time-sensitive, then the knowledge information in the conflict information group that is closest to the current time is determined as the target knowledge information. When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is authority priority, the knowledge information with the highest authority score among the information sources in the conflict information group is determined as the target knowledge information; the authority score is a score obtained by analyzing the reference network between information sources based on an authority score algorithm; When the conflict resolution method corresponding to the information domain to which the knowledge information in the conflict information group belongs is majority consensus assessment, the knowledge information in the conflict information group that appears more frequently than a frequency threshold is identified as the target knowledge information.

3. The method according to claim 1 or 2, characterized in that, The method further includes: When the multi-source knowledge information includes conflict information groups, conflict information is output. The conflict information includes at least one of the following: conflict type, conflict knowledge, and conflict level. The conflict type includes numerical conflict or logical conflict, and the conflict level is determined based on preset conflict level rules, which differ for different information domains.

4. The method according to claim 1 or 2, characterized in that, After receiving the input multi-source knowledge information, the method further includes: The multi-source knowledge information is preprocessed to obtain standardized knowledge information corresponding to each piece of knowledge information; A multi-dimensional credibility assessment is conducted on standardized knowledge information for each knowledge category to obtain credibility assessment results for each dimension. The overall credibility of the corresponding knowledge information is determined by comprehensively evaluating the credibility of each knowledge information across multiple dimensions. Information is filtered based on the overall credibility of various knowledge sources to obtain filtered multi-source knowledge information.

5. The method according to claim 4, characterized in that, The multi-dimensional credibility assessment includes at least two of the following: information source authority assessment, timeliness decay assessment, and knowledge consistency assessment. The standardized knowledge information is subjected to multi-dimensional credibility assessment to obtain credibility assessment results for each dimension, including: When the multi-dimensional credibility assessment includes an assessment of the authority of the information source, an authority score for the knowledge information is output based on the information source identifier in the standardized knowledge information and a pre-constructed authority graph. When the multi-dimensional credibility assessment includes a timeliness decay assessment, the timeliness score of the knowledge information is calculated based on the information release time and the decay factor of the information domain to which the knowledge information belongs in the standardized knowledge information. In the case where the multi-dimensional credibility assessment includes knowledge consistency assessment, a consistency score between knowledge information and known knowledge information is calculated using a graph attention network based on a pre-constructed knowledge graph, wherein the knowledge graph contains entity nodes, relation edges, and time information.

6. The method according to claim 5, characterized in that, The comprehensive credibility determination of corresponding knowledge information based on the credibility assessment results of various knowledge information across multiple dimensions includes: The overall credibility of knowledge information is obtained by weighted summation of the credibility assessment results from multiple dimensions; the weighting weights are different for different dimensions of credibility assessment.

7. The method according to claim 6, characterized in that, The method further includes: In response to receiving user feedback information based on the target knowledge information, the weighting of the multi-dimensional credibility assessment is adjusted based on the user feedback information.

8. The method according to claim 1, characterized in that, The method further includes: Non-conflicting knowledge information from the multi-source knowledge information is added to the knowledge base for large models to call and respond to questions.

9. An information processing device, characterized in that, The device includes: The information receiving module is used to receive input multi-source knowledge information, wherein the multi-source knowledge information includes knowledge information from at least two information sources; The conflict resolution module is used to resolve conflicts in the multi-source knowledge information group based on the information domain to which the knowledge information in the conflict information group belongs, when such conflict information group is contained in the multi-source knowledge information group, in order to obtain the target knowledge information; different information domains correspond to different conflict resolution methods; the conflict information group contains multiple knowledge information groups with different information sources and conflicting information; The first information output module is used to output the target knowledge information for the large model to call and respond to questions.

10. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one computer program, which is loaded and executed by the processor to implement the information processing method as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the information processing method as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, cause the computer to perform the information processing method as described in any one of claims 1 to 8.

Images