Multi-source information retrieval fusion method, device and equipment and readable storage medium

Abstract

The application relates to a multi-source information retrieval fusion method, device and equipment and a readable storage medium. The method comprises the following steps: analyzing acquired user query data, determining a query intention type, an intention representation and a constraint condition of the query data; determining a plurality of target information sources matched with the query intention type from a plurality of preset information sources; for each target information source, performing retrieval on the target information source according to the intention representation and the constraint condition to obtain information items corresponding to the intention representation; and performing fusion processing on the plurality of information items to generate a summarized answer. The method can improve the answer accuracy.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular to a multi-source information retrieval and fusion method, apparatus, device, and readable storage medium. Background Technology

[0002] With the development of artificial intelligence technology, large-scale AI models (such as ChatGPT and Tongyi 1000 Questions) have become a new direction in information processing due to their advantages in multi-domain question answering, reasoning, and dialogue generation. However, users are placing higher demands on the diversity, accuracy, and timeliness of information in various scenarios such as daily communication, work decision-making, and scientific research. For example, when faced with open questions, users expect to obtain general knowledge and data reasoning; regarding real-time information or dynamic content, user needs shift towards the latest news and data updates; and in specific fields or scientific research, users require more specialized content.

[0003] However, relying on large models to generate answers in related technologies cannot adapt to the needs of different scenarios, and the accuracy of the generated answers is low. Summary of the Invention

[0004] Therefore, it is necessary to provide a multi-source information retrieval and fusion method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of responses to the above-mentioned technical problems.

[0005] Firstly, this application provides a multi-source information retrieval and fusion method, including:

[0006] Analyze the acquired user query data to determine the query intent type, intent representation, and constraints of the query data;

[0007] Multiple target information sources matching the query intent type are determined from a set of preset information sources.

[0008] For each target information source, the target information source is retrieved according to the intent representation and the constraints to obtain information items corresponding to the intent representation;

[0009] Multiple information items are merged to generate a summary answer.

[0010] In one embodiment, determining multiple target information sources matching the query intent type from a preset set of multiple information sources includes:

[0011] Determine the quality feature data of each information source in the multiple information sources in different quality feature dimensions, as well as the demand preference data of the query intent type for the quality feature dimensions;

[0012] Based on the demand preference data and the quality characteristic data, determine the matching degree matrix between the query intent type and the multiple information sources;

[0013] The information sources corresponding to the preset values ​​of the elements in the matching degree matrix are identified as the target information sources that match the query intent type, thus obtaining multiple target information sources.

[0014] In one embodiment, the step of retrieving the target information source for each target information source based on the intent representation and the constraints to obtain an information item corresponding to the intent representation includes:

[0015] For each target information source, the target information source is retrieved according to the intent representation and the constraints to determine the original text that matches the intent representation;

[0016] If the original text is of the unstructured text type, then information extraction is performed on the original text to determine the information item corresponding to the intent representation from the original text.

[0017] In one embodiment, the step of fusing multiple information items to generate a summary answer includes:

[0018] Semantic alignment is performed on multiple information items to obtain the first candidate information item after semantic information alignment for each item.

[0019] For each of the first candidate information items, the confidence level of the first candidate information item in each preset evaluation dimension is determined according to a preset reliability evaluation model;

[0020] The composite confidence level of the first candidate information item is determined based on the confidence level.

[0021] Perform conflict detection on each of the first candidate information items to obtain conflict data between the first candidate information items;

[0022] Based on the composite confidence level and the conflict data, a corresponding conflict resolution strategy is generated.

[0023] Based on the first candidate information item, the confidence level, the composite confidence level, and the conflict handling strategy, a multi-source information fusion prompt is determined, and multiple first candidate information items are fused to generate a summary answer.

[0024] In one embodiment, generating a corresponding conflict resolution strategy based on the confidence level, the composite confidence level, and the conflict data includes:

[0025] If among the first candidate information items, there are multiple second candidate information items with a composite confidence level greater than or equal to the first preset confidence level, the generated conflict handling strategy includes sorting the multiple second candidate information items from largest to smallest according to their composite confidence levels, and determining the priority of each second candidate information item based on the sorting result.

[0026] And / or, if among the first candidate information items, there exists a third candidate information item with a composite confidence level greater than or equal to the second preset confidence level, and the conflict data among the multiple third candidate information items is non-conflicting, then the generated conflict handling strategy includes semantic fusion of the multiple first candidate information items to generate a neutral summary; the second preset confidence level is greater than the first preset confidence level.

[0027] And / or, if there are multiple sets of fourth candidate information items in the first candidate information items where the difference between any two of the composite confidence levels is within a preset range, and the conflict data between the fourth candidate information items is conflicting, then the generated conflict handling strategy includes displaying the conclusion, information source, and composite confidence level of each of the fourth candidate information items.

[0028] In one embodiment, the method further includes:

[0029] Determine the citation information of the semantic fragments in the summary answer, wherein the citation information includes at least one of the following: information source, original text, and attribute information associated with the original text;

[0030] Generate interactive references to the semantic fragments;

[0031] In response to a triggering operation for the interactive reference, the reference information referenced by the semantic fragment is displayed.

[0032] In one embodiment, determining the reference information of semantic segments in the summary answer includes:

[0033] The summary answer is split according to a preset semantic format to obtain semantic fragments;

[0034] For each semantic segment, the semantic segment and the original information segment in the original text corresponding to the semantic segment are encoded to obtain a first vector of the semantic segment and a second vector of each original information segment.

[0035] Based on the first vector and each of the second vectors, the similarity between the semantic segment and the original information segment is determined, and a preset number of target original information segments are determined from the original information segments based on the similarity.

[0036] The target original information fragment, the information source of the target original information fragment, and the attribute information associated with the target original information fragment are determined as the reference information of the semantic fragment in the summary answer.

[0037] Secondly, this application also provides a multi-source information retrieval and fusion device, comprising:

[0038] The data analysis module is used to analyze the acquired user query data to determine the query intent type, intent representation, and constraints of the query data.

[0039] The information source matching module is used to determine multiple target information sources that match the query intent type from a preset set of multiple information sources;

[0040] The retrieval module is used to retrieve information items corresponding to the intent representation for each target information source based on the intent representation and the constraints.

[0041] The information fusion module is used to fuse multiple information items to obtain fused information, summarize the fused information, and generate a summary answer.

[0042] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described above.

[0043] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0044] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0045] The aforementioned multi-source information retrieval and fusion method, apparatus, computer equipment, computer-readable storage medium, and computer program product identify user query data to determine the query intent type. Based on the query intent type, they determine multiple target information sources matching the query intent type from various information sources including different query channels. Then, they retrieve information items from each target information source using intent representation and constraints. By fusing and generatively summarizing information items from different query channels, they obtain a generative summary answer. This approach effectively combines and coordinates multiple heterogeneous data sources to meet the needs of different scenarios, improving the accuracy and interpretability of the answer. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a diagram illustrating the application environment of a multi-source information retrieval and fusion method in one embodiment.

[0048] Figure 2 This is a flowchart illustrating a multi-source information retrieval and fusion method in one embodiment;

[0049] Figure 3 This is a flowchart illustrating step 204 in one embodiment;

[0050] Figure 4 This is a flowchart illustrating a method for generating a summary response in one embodiment.

[0051] Figure 5 This is a flowchart illustrating a multi-source information retrieval and fusion method in another embodiment;

[0052] Figure 6 This is a block diagram of the dialogue system in one embodiment;

[0053] Figure 7 This is a structural block diagram of a multi-source information retrieval and fusion device in one embodiment;

[0054] Figure 8 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] With the development of artificial intelligence (AI) technology, large AI models (such as ChatGPT and Tongyi 1000 Questions) have become a new direction in information processing due to their advantages in multi-domain question answering, reasoning, and dialogue generation. Related technologies utilize large models to generate answers. However, while large models excel in general knowledge and natural language generation, they lack precision in providing information in specific domains. In particular, they lag behind in acquiring the latest events and dynamic information because their training data is limited and update cycles are long, leading to outdated content. Furthermore, large models are expensive to train and have limited scalability. In other words, answers generated by large models cannot meet the needs of real-time information and domain-specific knowledge, resulting in low accuracy.

[0057] However, data sources other than large-scale model-generated answers, such as web search, public libraries, personal domain libraries, and external data interfaces, cannot adapt to different scenarios in answer generation to ensure accuracy. For example, the quality of results returned by web search is unstable, and its coverage of professional fields is insufficient, with a lot of noise, requiring further filtering and verification of information. Public libraries update data slowly and lack the ability to respond to cutting-edge or dynamic information. Personal domain libraries have limited knowledge scope and cannot meet general or cross-domain needs. External data interfaces (such as financial data APIs) can provide structured and real-time data, suitable for querying professional information or numerical data. However, these data interfaces are usually limited by call frequency, are costly, and the returned data lacks context, making them unsuitable for direct use in generating complex answers.

[0058] Therefore, in response to the technical problem of how to improve the adaptability and accuracy of the generated answers, a multi-source information retrieval fusion method is proposed.

[0059] The multi-source information retrieval and fusion method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. The terminal analyzes the acquired user query data to determine the query intent type, intent representation, and constraints; it identifies multiple target information sources matching the query intent type from a preset pool of information sources; for each target information source, it retrieves information items corresponding to the intent representation based on the intent representation and constraints; it merges multiple information items to obtain fused information, summarizes the fused information, and generates a summary answer.

[0060] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, etc. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0061] In one exemplary embodiment, such as Figure 2 As shown, a multi-source information retrieval and fusion method is provided, which is applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps 202 to 208. Wherein:

[0062] Step 202: Analyze the acquired user query data to determine the query intent type, intent representation, and constraints of the query data.

[0063] The query intent type can include, but is not limited to, factual, comparative, and causal types. The query intent type can be constructed by pre-analyzing potential user needs, classifying and defining possible user question types, and building a predefined intent set to facilitate subsequent matching of user questions.

[0064] For example, query intent types can include knowledge and fact-based queries (e.g., common sense, definition / explanation questions, causal queries, application queries, hypothesis queries), logical reasoning queries, task-based queries or process-based queries, conversational queries (e.g., casual conversation / social / psychological counseling / emotional expression queries), real-time information queries, predictive queries, tutorial / guidance queries, personalized information queries, data queries and analysis, and queries specifying sources. It is understandable that user query data can include at least one query intent type, and each query intent type can have its own corresponding intent representation and constraints.

[0065] Intent representation, or key entities, can be understood as entities representing user intent. Key entities help the system determine the focus of a user's query, thereby enabling more accurate location of relevant information sources and data during information retrieval. Furthermore, key entities can refer to core nouns or noun phrases in a user's query, typically representing the main objects or topics of interest to the user. These entities are fundamental to understanding user intent and performing information retrieval tasks. For example, key entities can be objects (e.g., product names, place names, etc.), concepts (e.g., artificial intelligence and weather changes), and events.

[0066] Constraints are specific requirements or restrictions specified by users in their queries, directly impacting the filtering and ranking of search results. Examples include time constraints (e.g., last month's report, data from the last three months), location constraints (e.g., progress in region A), type constraints (e.g., academic papers), and subject constraints (e.g., technical challenges). Constraints help the system further refine the search scope, ensuring that the returned information better matches the user's specific needs and context. By combining key entities and constraints, the system can perform information retrieval tasks more precisely, providing users with more relevant and accurate information.

[0067] For example, intent recognition is performed on user query data based on a natural language processing model. This identifies the type of query intent and extracts key entities and constraints from the user query data. These key entities and constraints provide a semantic foundation for subsequent information retrieval. Intent recognition can be achieved through fine-tuning of a generative large model or some traditional algorithms, which will not be elaborated upon here.

[0068] Step 204: Determine multiple target information sources that match the query intent type from the preset multiple information sources.

[0069] The various information sources include multiple heterogeneous data sources, such as large models, web search, public libraries, personal domain libraries, and external data interfaces. Web search excels at retrieving real-time internet content and is suitable for providing the latest news, trending events, and dynamic data. Public libraries, such as Wikipedia and open research databases, can provide authoritative, structured academic knowledge, historical data, and scientific facts, making them suitable for providing background information. Personal domain libraries can meet users' individual needs, providing highly relevant and customized information. External data interfaces (such as financial data APIs) can provide structured and real-time data, suitable for querying professional information or numerical data. The correspondence between query intent types and target information sources is predetermined or can be determined by the matching degree between query intent types and various information sources.

[0070] For example, taking a scientific research scenario, when using a dialogue system, it will conduct open-ended questions and answers on professional domain knowledge, as well as some everyday questions, and also need to handle different task types in scientific research scenarios. The corresponding target information source is determined based on each intent type in the scientific research scenario.

[0071] For example, for query intent types categorized as knowledge and facts (common sense, definition / explanation questions, causal questions, applications, hypotheses), if the query is defined as a user asking about general common sense, involving general knowledge and popular topics, such as "What is the tallest mountain in the world?" or "Why is the sky blue?", then the confirmed target information sources could be large models, public libraries, and personal domain libraries. If the query is defined as a user wanting the system to explain a term, concept, or principle, such as "What is quantum computing?" or "Explain how artificial neural networks work," then the confirmed target information sources could be large models, public libraries, and personal domain libraries. If the query is defined as a user asking a professional question about a specific field, usually involving in-depth knowledge or terminology, such as "What is the difference between machine learning and deep learning?" or "How do antibiotics inhibit bacterial growth?", then the confirmed target information sources could be large models, public libraries, and personal domain libraries.

[0072] If defined as users wanting to know the cause or consequence of an event or phenomenon, such as "Why do plants need sunlight to grow?" or "What are the effects of not exercising?", then the identified target information sources could be large models, public libraries, and personal domain libraries. If defined as users proposing hypothetical scenarios to explore the possible outcomes of "what if" situations, such as "How long can a fish survive if it leaves the water?", then the identified target information sources could be large models, public libraries, and personal domain libraries.

[0073] For query intent types that are logical reasoning questions, defined as complex problems requiring reasoning or multi-step calculations, such as "If a tortoise crawls at a speed of 5 kilometers per hour, how long will it take it to crawl 20 kilometers?" or "If A is greater than B, B is smaller than C, and C is greater than D, then which is the smallest?", the identified target information source could be a large model. For query intent types that are task-oriented questions or process operations, defined as user requests to complete a task or instructive operation involving a specific workflow, such as translation, writing, coding, drawing, booking a restaurant, setting reminders, or making settings, for example, "Please set a reminder for a meeting tomorrow at 3 PM" or "Open the lighting settings page," then the identified target information source could be an external application programming interface (API).

[0074] For query intent types that are conversational (casual conversation / social interaction / psychological counseling / emotional expression), if defined as users engaging in informal small talk or expressing emotions, there is usually no explicit knowledge requirement. For example, "Do you like watching movies?", "I'm a little tired today," or "How are you?", then the confirmed target information source model is particularly adept at handling small talk and social conversations, generating natural dialogues. If defined as users expressing emotions, feelings, or seeking emotional support, for example, "I'm feeling down today, can you comfort me?", or "I feel a bit lost about the future, how should I adjust my mindset?", then the confirmed target information source, i.e., the large model, can generate appropriate emotional support or comfort based on the user's emotions.

[0075] For queries targeting real-time information, defined as users wanting to obtain currently occurring events or instant information—such as "What's the weather like today?" or "Can you summarize last week's news headlines?"—then the identified target information source is the large model: this model can summarize information from multiple sources and generate a concise summary answer. Web search can obtain real-time information such as news, weather, and exchange rates. External application programming interfaces (APIs), such as weather APIs and financial data APIs, can obtain real-time data.

[0076] For query intent types that are predictive, such as "What are the development trends of artificial intelligence in the next ten years?", the identified source of information is a large model: a model that can make reasonable predictions based on existing trends and historical data. Web search can be used to retrieve predictive reports or trend analyses in relevant fields.

[0077] For queries with the intent type "tutorial / guide," if the user is defined as wanting step-by-step instructions or learning guidance—for example, "How do I install Python?", "What are the steps to make pasta?", "How do I draw?", or "How do I learn to code?"—then the identified target information source, the broad model, is one that provides detailed step-by-step instructions or learning guidance. Web search: This provides readily available tutorials or step-by-step guides, especially for learning tools and skills.

[0078] For queries targeting personalized information, if a user desires answers based on their personal data or customized information, access to a user-specific database is typically required. For example, queries like "What were my to-do lists last week?" or "What time is my doctor's appointment?" identify the target information source as the large-scale model. The system should be able to summarize various information sources and generate a clean, concise summary answer. External application programming interfaces (APIs) are also available, such as database APIs for querying user personal data like schedules and task lists.

[0079] For query intent types defined as data query and analysis, if it's defined as the user wanting the system to query and analyze certain data—for example, "Analyze my sales data from the past five years," or "Show me the profit distribution for the first quarter of this year"—then the confirmed target information source is statistical data in the relevant field retrieved through web search. External application programming interfaces (APIs): For example, using database APIs or data analysis interfaces to provide data query and analysis results.

[0080] For query intent types specifying a source, if it's defined as the user explicitly indicating where they want to obtain information, such as "Please search online for the answer to: How is the iPhone 15 phone reviewed?", then the confirmed target information source is web search, retrieving statistical data in the relevant field. For example, "Please search my personal document library for the answer to the question: The causes of earthquakes," then the confirmed target information source is a personal domain library: if the user has uploaded relevant domain knowledge bases, they can provide content. For example, "Please search wikis or authoritative open-source literature for the answer to the question: The causes of earthquakes," then the confirmed target information source is public libraries: these can serve as a basis for information aggregation, providing authoritative background information. For example, "Please have a large model directly answer the question: The causes of earthquakes," then the confirmed target information source is a large model: directly answering the user's provided information.

[0081] Step 206: For each target information source, retrieve the target information source according to the intent representation and constraints to obtain the information item corresponding to the intent representation.

[0082] The retrieval methods can be, but are not limited to, using RAG (Retrieval-Augmented Generation) technology, web retrieval, API calls, and other retrieval techniques to search for multi-source target information. Examples include full-text search based on keyword inverse indexes, semantic search based on sentence vectors, or field queries based on structured data. Search engines can include open-source or commercial products such as Elasticsearch, Solr, and Milvus, and can also be deployed in local or cloud environments in conjunction with custom recall logic.

[0083] Furthermore, during the process of calling external data sources or knowledge services, standard interface protocols (such as RESTful API, GraphQL, and plugin protocols) can be used to connect with and interact with third-party platforms, including but not limited to public databases and file services. The encapsulation and parsing of interface calls can be implemented using existing service scheduling frameworks or middleware. Those skilled in the art can implement this based on conventional network programming methods, which will not be elaborated here.

[0084] Optionally, the target information source is retrieved based on the intent representation and constraints to obtain the corresponding original text. Information extraction is then performed on the original text to obtain information items corresponding to the intent representation. The method for retrieving the original text can be implemented using existing methods and will not be elaborated here. Information extraction can be performed by an information extraction model (such as BERT-NER) that converts the original text into structured knowledge units, and then extracts information items corresponding to the intent representation from these structured knowledge units.

[0085] Step 208: Merge multiple information items to obtain merged information and generate a summary answer.

[0086] The fusion process can be achieved through a generative model, which generates a comprehensive, summarizing answer based on information items retrieved from multiple information sources, rather than simply piecing together or listing different pieces of information. Furthermore, before fusing multi-source information, it is necessary to ensure that the information from different sources is semantically consistent, meaning it expresses the same facts or viewpoints.

[0087] Optionally, when user input contains multiple questions, the system typically needs to break these sub-questions down and then retrieve and process each sub-question independently. The final result is like a comprehensive integration of content retrieved from multiple data sources, rather than a simple concatenation and output. The splitting method can be implemented using existing techniques, which will not be elaborated upon here.

[0088] For example, multiple information items can be fused to obtain fused information. To generate a summary answer, the multiple information items can be semantically aligned to identify the same object in different expressions, which facilitates subsequent confidence comparison and conflict detection. Then, confidence assessment and conflict detection are performed, and a conflict resolution strategy, i.e. a conflict resolution strategy, is determined based on the structure of the conflict detection. The results of semantic alignment, confidence assessment, and conflict detection, as well as the conflict resolution strategy, are used as prompt rules when generating the generative model to constrain the generation of the final answer, thereby achieving information fusion and generative answer.

[0089] The aforementioned multi-source information retrieval and fusion method identifies user query data to determine the query intent type. Based on the query intent type, it identifies multiple target information sources matching the query intent type from various information sources including different query channels. Then, it uses intent representation and constraints to retrieve information items from each target information source. By fusing and generatively summarizing the information items from different query channels, it obtains a generative summary answer. This approach effectively combines and coordinates multiple heterogeneous data sources to meet the needs of different scenarios and improves the accuracy and interpretability of the answer.

[0090] The following provides an implementation method for identifying multiple target information sources that match a query intent type. In an exemplary embodiment, such as... Figure 3 As shown, step 204 includes steps 302 to 306. Wherein:

[0091] Step 302: Determine the quality feature data of each information source in the multiple information sources in different quality feature dimensions, as well as the demand preference data of query intent types for quality feature dimensions.

[0092] The different quality feature dimensions are determined according to actual needs. Feature dimensions may include attribute feature dimensions such as timeliness, authority, personalization and content depth.

[0093] Demand preference data can include users' questions about various general knowledge, covering common knowledge and popular topics; requests for system explanations of a certain term, concept or principle; answers to professional questions about a specific field; requests to know the causes or consequences of events or phenomena; proposals of hypothetical scenarios to explore possible outcomes of "what if" situations; requests for reasoning or multi-step calculations on complex problems; and requests to complete a certain task or instructional operation involving specific operational procedures.

[0094] Step 304: Based on demand preference data and quality characteristic data, determine the matching degree matrix between query intent type and multiple information sources.

[0095] For example, query intent type preference data for quality feature dimensions is matched with multiple information sources to obtain a matching degree matrix between query intent type and multiple information sources. For instance, depending on the specific research scenario, there are large models, web search, public repositories, personal literature databases, and external data APIs. These data sources themselves have significant differences in attributes such as timeliness, authority, personalization, and content depth. Based on the identified 10 intents, the matching degree of five data sources will be evaluated separately, focusing on the different needs of different intents for information timeliness, authority, personalization, and content depth, thereby determining the matching degree of the five data sources for these different intents (0 for no match, 1 for match), as shown in the information source matching matrix in Table 1 below. A matching degree of 0 will not participate in this round of information retrieval, and the target information source matched for each query intent type will be obtained.

[0096] Table 1 Information Source Matching Matrix

[0097]

[0098] Based on the matching method described above, it can be seen that when the system handles different user problems, it can adaptively match different information data sources for different problems. For other systems or systems with different requirements for information sources, the values ​​in the table can be updated to flexibly adapt to various user needs, obtain the corresponding target information source, and improve the accuracy and efficiency of data source retrieval.

[0099] Step 306: The information sources corresponding to the preset values ​​of the elements in the matching degree matrix are identified as the target information sources that match the query intent type, thus obtaining multiple target information sources.

[0100] The preset value can be 1.

[0101] In this embodiment, the priority of the information source is dynamically adjusted according to the matching degree between the user's intent and the information source, which can adapt to the differentiated needs for information timeliness, authority, and personalization in different scenarios.

[0102] To further extract structured data from raw text that is easier for machines to process, in an exemplary embodiment, for each target information source, the target information source is retrieved based on intent representation and constraints to obtain information items corresponding to the intent representation, including:

[0103] For each target information source, the target information source is retrieved based on the intent representation and constraints to determine the original text that matches the intent representation; if the type of the original text is unstructured text, information extraction is performed on the original text to determine the information items corresponding to the intent representation from the original text.

[0104] For example, if the original text obtained is not structured text, an information extraction model is needed to extract it, such as BERT-NER, to convert the original text of the retrieved information items into structured knowledge units and determine the information items corresponding to the intent representation. If the original text is structured text, no conversion is needed. This approach can extract useful information from unstructured text and transform it into a structured form that is easier for machines to process.

[0105] In one exemplary embodiment, a method for generating summary-style answers is provided, such as... Figure 4 As shown, it includes the following steps:

[0106] Step 402: Semantically align multiple information items to obtain the first candidate information item after semantic information alignment for each item.

[0107] Semantic alignment can be achieved by clustering "synonymous facts" expressed in different ways. Even if the information is expressed differently, if they represent the same fact, they will be clustered together in the vector space. For example, "the Moho discontinuity is about 66 kilometers thick" and "the Moho interface is 66 km deep" can be considered a pair of synonymous sentences. This can be implemented using semantic vector models such as SBERT (Sentence-BERT) and SimCSE (Simple Contrastive Sentence Embeddings). Specific implementation methods are available and will not be elaborated upon here.

[0108] It should be noted that semantic alignment is necessary to identify the same object in different expressions, which facilitates subsequent confidence comparison and conflict detection.

[0109] Step 404: For each first candidate information item, determine the confidence level of the first candidate information item in each preset evaluation dimension according to the preset reliability evaluation model.

[0110] Step 406: Determine the composite confidence level of the first candidate information item based on the confidence level.

[0111] The pre-set reliability assessment model includes at least the following assessment dimensions: source authority, data timeliness, methodological justification, and intent matching.

[0112] The source score measures the credibility of the information source, prioritizing academic / official channels, etc. Its scoring range is 0-1, with a weight of [missing value]. Specifically, this includes: national authoritative institutions (such as CENC and NASA) with a score of 1.00; SCI journal papers (Q1 / Q2) with a score of 0.90; high-quality preprints (such as arXiv and bioRxiv) with a score of 0.80; university / institute project reports or databases with a score of 0.70; media reports, Wikipedia, and other secondary sources with a score of 0.50; and community Q&A / non-professional forums (such as Zhihu and Baidu) with a score of 0.30.

[0113] The time score ensures that relatively recent data / research is used, which is particularly important in rapidly evolving fields (such as earthquake monitoring and climate forecasting). The corresponding scoring range is 0-1, with a weight of [missing value]. Specifically, the following criteria apply: a recommendation score of 1.00 is given for publications within 6 months of the current date; 0.85 for publications within 1 year; 0.70 for publications within 2-3 years; 0.40 for publications exceeding 5 years; and 0.30 for publications exceeding 10 years.

[0114] The method score measures whether the data acquisition, experimentation, and modeling behind the information are detailed, reasonable, and reliable. Its recommended score ranges from 0 to 1, with a weight of [weight missing]. Specifically, the quality of the method description includes: clearly stating the data source, method, parameters, etc., with a recommended score of 1; providing a complete method or experimental procedure, with a recommended score of 0.90; briefly describing the method but lacking details, with a recommended score of 0.70; providing no method description, only giving results or conclusions, with a recommended score of 0.40; and relying on speculative content or verbal statements, with a recommended score of 0.20.

[0115] The intent score determines whether the information truly answers the core of the user's question; it has a score range of 0-1 and a weight of [missing value]. For example, sentence vectors (such as SBERT) can be used to calculate the semantic similarity between text and user questions. Specifically, a similarity range of ≥0.9 corresponds to a matching score of 1.00; a similarity range of 0.8-0.9 corresponds to a matching score of 0.90; a similarity range of 0.7-0.8 corresponds to a matching score of 0.75; a similarity range of 0.6-0.7 corresponds to a matching score of 0.60; and a similarity range of <0.6 corresponds to a matching score of 0.40.

[0116] Based on this, the composite confidence score (confidence_score) is calculated using a linear weighting method:

[0117] ,

[0118] Among them, the default weight (e.g., It can be modified according to actual needs and scenarios.

[0119] Step 408: Perform conflict detection on each first candidate information item to obtain conflict data between the first candidate information items.

[0120] Among them, conflict detection can detect the numerical, qualitative, and causal aspects of information items. For example, inconsistent numerical values, contradictory qualitative assertions, and mutually exclusive causal inferences are considered as conflicts.

[0121] For example, common conflict types can include: numerical conflicts: differences between the same indicator exceeding ±5% (e.g., "66km" vs. "70km"), qualitative conflicts: mutually exclusive conclusions (e.g., "plate stability" vs. "risk of fracture"), and causal conflicts: contradictory reasoning chains (e.g., A→B vs. A→¬B). Optionally, specific conflict detection techniques are as follows: For numerical conflicts, the numerical values ​​and units in the text can be extracted using regular expressions, standardized, and the relative percentage difference calculated. When the numerical difference between the same indicator exceeds a preset threshold (default ±5%, adjustable according to the domain), it is considered a conflict. Data confidence is also considered to avoid misjudgment; for example, the relative difference between "66km" and "70km" is 6.06%, exceeding the threshold, thus marking it as a numerical conflict.

[0122] For qualitative conflicts, text vector similarity is calculated using a pre-trained language model (such as BERT), and mutually exclusive expressions (such as the semantic opposition between "stable" and "fracture") are identified by combining an antonym lexicon.

[0123] For causal conflicts, dependency parsing is used to extract the subject-verb-object structure, and logical rules are used to verify inference contradictions (such as detecting the mutual exclusion relationship between "A causes B" and "A does not cause B"). The conflict detection results should output conflict data including conflict type, conflict text fragments, and confidence levels for each preset evaluation dimension, which can be used as input for subsequent integration and processing as prompt content.

[0124] Step 410: Generate the corresponding conflict resolution strategy based on the composite confidence level and conflict data.

[0125] The conflict resolution strategy can be as follows: based on the composite confidence level and the first pre-set confidence level, selectively retain information items with a confidence level greater than or equal to the first pre-set confidence level, and discard information items with a confidence level less than the first pre-set confidence level; when multiple third candidate information items with a composite confidence level greater than or equal to the second pre-set confidence level coexist, a neutral summary answer will be generated together; if there are multiple fourth candidate information items among the first candidate information items whose difference between any two composite confidence levels is within a preset range, but the conclusions are contradictory (such as inconsistent numerical values, contradictory qualitative assertions, mutually exclusive causal inferences, etc.), multiple conclusions and sources will be displayed for the user to judge. The conflict resolution strategy will also serve as a prompting rule during model generative model generation, acting as input for multi-source information fusion prompts to constrain the generation of the final answer.

[0126] Furthermore, based on the composite confidence level and conflict data, a corresponding conflict resolution strategy is generated, including at least one or more of the following three cases:

[0127] Scenario 1: If there are multiple second candidate information items among the first candidate information items with a composite confidence level greater than or equal to the first preset confidence level, the generated conflict handling strategy includes sorting the multiple second candidate information items from largest to smallest according to their composite confidence levels, and determining the priority of each second candidate information item based on the sorting results.

[0128] The first preset confidence level can be set according to actual needs, for example, it can be 0.3. When the confidence level of information is too low (<0.3), the information is considered unreliable and can be ignored and not used.

[0129] For example, if there are multiple second candidate information items among the first candidate information items with a composite confidence level greater than or equal to the first preset confidence level, the generated conflict handling strategy includes, in the information fusion and generative response stage, sorting the composite confidence levels of the multiple second candidate information items from largest to smallest, determining the priority of each second candidate information item based on the sorting result, and using the expression with the highest priority, that is, the highest composite confidence level, as the main basis for the generative response.

[0130] Scenario 2: If there are multiple third candidate information items in the first candidate information item with a composite confidence level greater than or equal to the second preset confidence level, and there is no conflict data between the multiple third candidate information items, then the generated conflict handling strategy includes semantic fusion of the multiple third candidate information items to generate a neutral summary; the second preset confidence level is greater than the first preset confidence level.

[0131] The second preset reliability can be set according to actual needs, for example, it can be 0.75. It can be understood that, based on Case 1, after discarding candidate information items with a reliability lower than the first preset reliability from the first candidate information items, if Case 2 still exists, it will be processed in conjunction with the processing method of Case 2.

[0132] Scenario 3: If there are multiple fourth candidate information items in the first candidate information items whose difference between any two composite confidence levels is within a preset range, and the conflicting data between the fourth candidate information items is conflicting, then the generated conflict handling strategy includes displaying the conclusion, information source, and composite confidence level of each fourth candidate information item.

[0133] The preset range can be less than 0.1. Conflicts can manifest as one or more of the following: qualitative contradiction, causal mutual exclusion, and high numerical deviation.

[0134] Understandably, Case 3 is based on Case 1. After discarding candidate information items with a lower than the first preset confidence level from the first candidate information items, if Case 3 exists, it will be processed in conjunction with the processing method of Case 3.

[0135] Step 412: Determine multi-source information fusion prompts based on the first candidate information items, confidence level, composite confidence level, and conflict handling strategy; fuse multiple first candidate information items to generate a summary answer.

[0136] The first candidate information item can be displayed as a list of information items or in JSON (JavaScript Object Notation, JSON format). For example, the information item list can be filled with structured information items, such as JSON or list format, each item including: information content, source (e.g., web network / large model / public library / personal local library / API interface), confidence level, composite confidence level, and whether there are qualitative conflicts or numerical differences, etc. Optionally, the multi-source information fusion prompt can be as follows:

[0137] You are a researcher in a specialized field, integrating search results from different sources to generate an authoritative, credible, and well-structured research summary. The following are multi-source information items (including their confidence scores and source information). Please handle conflicts and write a summary response according to the following rules:

[0138] [Explanation of Conflict Resolution Strategies]:

[0139] 1. Confidence level selection strategy:

[0140] When the confidence level of an information item is too low (i.e. less than the first preset confidence level of 0.3), the information item is considered unreliable and should be ignored and not used. During the generation process, the second candidate information item with a higher composite confidence level will be given priority and its description will be regarded as the main basis.

[0141] 2. Strategy for coexistence of multiple high-confidence information:

[0142] If there are multiple information items with high confidence (i.e., greater than or equal to the second pre-set confidence level of 0.75) but without conflict, i.e., the third candidate information items, please integrate their key points and summarize them in neutral language.

[0143] Be sure to retain the strengths of each viewpoint to ensure that your response is comprehensive and representative.

[0144] 3. Strategies involving similar confidence levels but contradictory conclusions (qualitative contradictions, mutually exclusive causal relationships, and significant numerical bias):

[0145] If multiple confidence levels are similar (i.e., the difference between any two composite confidence levels is <0.1) and there is a substantial conflict in the conclusions, i.e., the fourth candidate information item, please list each conclusion, its source and confidence level for the user to make further judgments.

[0146] You can use expressions such as "There are differences in research perspectives" or "Some studies believe that... while others point out that..." to show the differences in viewpoints and demonstrate the ability to trace the source.

[0147] Please complete the following tasks:

[0148] Based on the information provided below, generate a summary research response;

[0149] Follow the above three information processing rules;

[0150] The style of the response should be professional, neutral, and verifiable;

[0151] The summary should mention the sources of different conclusions (indicated by literature or data sources);

[0152] [List of Information Items]:

[0153] <Enter structured information items here, such as JSON or list format. Each item includes:>

[0154] -Information Content

[0155] - Sources (web networks / large models / public libraries / personal local libraries / API interfaces)

[0156] - Confidence score

[0157] - Are there any qualitative conflicts or numerical differences?

[0158] >

[0159] In the above embodiments, information items are scored from multiple dimensions such as source authority, data timeliness, methodological justification, and semantic matching with user intent. This can more effectively handle information conflicts and uncertainties, and can dynamically adjust the information fusion method according to the user's query intent and the characteristics of the information source, thereby improving the reliability and credibility of the answer.

[0160] To ensure the traceability and interpretability of generative responses, after completing the fusion of multi-source information and the generation of summary responses, a semantic mapping relationship is constructed between the generated content and the original information to build a traceable question-and-answer response framework.

[0161] In one example embodiment, by determining the reference information of a semantic fragment in a summary response, the reference information includes at least one of the following: information source, original text, and attribute information associated with the original text; generating an interactive reference to the semantic fragment; and displaying the reference information referenced by the semantic fragment in response to a triggering operation on the interactive reference.

[0162] Semantic fragments can be derived by breaking down a summary answer into sentences or semantic clauses. Semantic clause level refers to semantic analysis of text from the perspective of clauses in natural language processing. It considers the semantic structure within clauses and the logical relationships between clauses. Attribute information includes publication time and paragraph position in the corresponding original text. Interactive citations can be, but are not limited to, numbers or links. Optionally, when a user clicks on an interactive citation, information such as the original text fragment cited, source name, and publication time are displayed, supporting structured output of source information for use by the front-end interface or API system.

[0163] Optionally, determining the reference information of semantic segments in the summary answer includes: splitting the summary answer according to a preset semantic format to obtain semantic segments; for each semantic segment, encoding the semantic segment and the original information segments in the original text corresponding to the semantic segment to obtain a first vector of the semantic segment and a second vector of each original information segment; determining the similarity between the semantic segment and the original information segments based on the first vector and each second vector; determining a preset number of target original information segments from the original information segments based on the similarity; and determining the target original information segments, the information source of the target original information segments, and the attribute information associated with the target original information segments as the reference information of the semantic segments in the summary answer.

[0164] For example, the summary response is split at the sentence or semantic clause level to obtain the generated sentences in the summary response; each generated sentence and the information fragment obtained from the original retrieval are uniformly encoded into sentence vectors; the encoding method can be BERT, SBERT, SimCSE, or the embedding output of a locally deployed large language model. The top N sentences with the highest relevance to the original information source for each sentence in the summary response are calculated using cosine similarity, and a mapping relationship is established. When the mouse hovers over a sentence in the response, clicking on each [number] expands to display a "Source Pop-up," showing relevant fragments, links, and the location of the original paragraph.

[0165] In the above embodiments, by supporting the source tracing method at the semantic fragment dimension, the system records the source information, original content fragments, and other source information of the generated answer, allowing users to click to view the original evidence behind the answer, thus ensuring the interpretability and transparency of the system.

[0166] In one exemplary embodiment, such as Figure 5 As shown, a multi-source information retrieval and fusion method is provided, which is applied to... Figure 1 Taking the terminal in the example, the explanation includes the following steps:

[0167] Step 502: Analyze the acquired user query data to determine the query intent type, intent representation, and constraints of the query data.

[0168] Step 504: Determine multiple target information sources that match the query intent type from the preset multiple information sources.

[0169] Step 506: For each target information source, retrieve the target information source according to the intent representation and constraints to obtain the information item corresponding to the intent representation.

[0170] Step 508: Merge multiple information items to generate a summary answer.

[0171] Step 510: Determine the reference information of the semantic fragments in the summary response. The reference information includes at least one of the following: information source, original text, and attribute information associated with the original text.

[0172] Step 512: Generate an interactive reference to the semantic fragment, and in response to a triggering operation on the interactive reference, display the reference information referenced by the semantic fragment.

[0173] It should be noted that the implementation method of this embodiment can be achieved in the manner described above, and will not be elaborated here.

[0174] The following provides a dialogue system based on the aforementioned multi-source information retrieval and fusion method. The dialogue system includes a user input module, a natural language understanding module, a dialogue management module, a multi-source retrieval module, an information fusion and tracing module, and a system output module, such as... Figure 6 As shown, each module is responsible for different stages from user input to response generation, ensuring that the system can ultimately achieve an "AI-generated but explainable" scientific research question-and-answer system.

[0175] Users can engage in question-and-answer dialogues with the system through the user dialogue interface provided on the system's PC client. User input, i.e., the data queried, can include only the most recently entered information, several recently entered information, or all historical information entered by the user; there are no restrictions on this.

[0176] After receiving user input, the user input is passed to the Natural Language Understanding (NLU) module. This module is responsible for parsing the user input, using a natural language processing model to perform semantic parsing on the user input query, building an intent classifier to identify the type of query intent (such as factual, comparative, causal, etc.), extracting key entities, constraints, and query context in the query, and finally determining the user's intent.

[0177] The output of the NLU module is sent to the dialogue management module, which guides the user to the multi-source information retrieval module based on the identified intent and updates the dialogue status synchronously. The multi-source information retrieval module evaluates the matching degree between various predefined information sources and the user's intent to determine which information sources to invoke. It combines RAG technology with various retrieval techniques, including web retrieval and API calls, to retrieve information from multiple sources.

[0178] After the search is completed, the system enters the information fusion and tracing module, which performs consistency checks and conflict resolution on information items retrieved from multiple sources. Based on the search results, it generates appropriate language responses using natural language generation (NLG) technology, provides feedback to the user, and uses a generative language model to fuse information to generate a summary answer. It also records tracing information such as the source of information and original content fragments in the answer.

[0179] The generated response is sent back to the user through the user dialogue interface or API provided by the system's PC. The user can continue to have multiple rounds of dialogue based on the feedback, and the system repeats the above steps.

[0180] In another exemplary embodiment, the user sends natural language input to the system via a chat interface or API, allowing the user to input complex questions simultaneously in a single-turn conversation. For example, the user inputs: "Please query the isotopic data of granite from the OnePetrology database and analyze the formation of granite using the data." The user needs to first retrieve the isotopic data of granite through the database API interface, and then analyze the formation of granite using the data. Therefore, the user intent is first broken down into two sub-intents, namely... Figure 1 Please query the isotopic data of granite from the OnePetrology database. Figure 2 Based on data analysis of the formation of granite, the corresponding intent types are data acquisition intent and professional analysis and reasoning intent. The target information source corresponding to the data acquisition intent is the specified source - calling the interface, while the target information source corresponding to the professional analysis and reasoning intent is the target information source corresponding to knowledge and facts (common sense, definition / explanation questions, causal questions, applications, hypotheses).

[0181] Based on the matching results of the target information source, for the intention Figure 1 We will meet user needs by directly accessing the OnePetrology database connected to the system. For [the specific user's needs]... Figure 2 We will obtain relevant information items by calling large models, web search, public libraries, and personal local paper repositories.

[0182] After obtaining information items through the following different information channels, the system then uses our confidence score to obtain the confidence scores for different information items, as shown in Table 2:

[0183] Table 2 Confidence scores for different information items

[0184]

[0185] In the "0.25×" column, "×" indicates that the result is less than a preset threshold and can be discarded. By integrating the above information items using the multi-source information fusion general Prompt, we will generate the final answer.

[0186] In the above embodiments, relevant information is retrieved from multiple heterogeneous information sources through intent recognition; confidence scores are assigned to each information item based on source authority, data timeliness, methodological justification, and intent matching; semantic alignment and consistency checks are performed, conflicting data is identified and processed using a fusion strategy; a comprehensive and authoritative answer is generated by calling a large language model based on a summary prompt input; and a source tracing mechanism is used to associate the generated content with the original information items and their sources, supporting source tracing. In other words, by constructing a system of "multi-source fusion + confidence assessment + interpretable generation," the capabilities of the intelligent question-answering system in multi-source data fusion, consistency processing, and result interpretability can be significantly improved. It can handle various types of user needs, such as open-ended questions, real-time information queries, and predictive questions, exhibiting stronger versatility and adaptability. Simultaneously, it provides users with more transparent and traceable answers, significantly enhancing the intelligent question-answering system's capabilities in multi-source data fusion and result interpretability.

[0187] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0188] Based on the same inventive concept, this application also provides a multi-source information retrieval and fusion apparatus for implementing the multi-source information retrieval and fusion method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more embodiments of the multi-source information retrieval and fusion apparatus provided below can be found in the limitations of the multi-source information retrieval and fusion method described above, and will not be repeated here.

[0189] In one exemplary embodiment, such as Figure 7 As shown, a multi-source information retrieval and fusion device is provided, including: a data analysis module 702, an information source matching module 704, a retrieval module 706, and an information fusion module 708, wherein:

[0190] The data analysis module 702 is used to analyze the acquired user query data to determine the query intent type, intent representation, and constraints of the query data.

[0191] The information source matching module 704 is used to determine multiple target information sources that match the query intent type from a set of preset information sources.

[0192] The retrieval module 706 is used to retrieve information items corresponding to the intent representation for each target information source based on the intent representation and constraints.

[0193] The information fusion module 708 is used to fuse multiple information items and generate a summary answer.

[0194] The aforementioned multi-source information retrieval and fusion device identifies the user's query data to determine the query intent type. Based on the query intent type, it identifies multiple target information sources that match the query intent type from various information sources, including different query channels. Then, it uses intent representation and constraints to retrieve information items from each target information source. By fusing and generatively summarizing the information items from different query channels, it obtains a generative summary answer. This method effectively combines and coordinates multiple heterogeneous data sources to meet the needs of different scenarios and improves the accuracy and interpretability of the answer.

[0195] In an exemplary embodiment, the information source matching module 704 is used to determine the quality feature data of each information source in different quality feature dimensions among multiple types of information sources, as well as the demand preference data of query intent type for quality feature dimensions.

[0196] Based on demand preference data and quality characteristic data, determine the matching degree matrix between query intent type and multiple information sources;

[0197] The information sources corresponding to the preset values ​​of the elements in the matching degree matrix are identified as the target information sources that match the query intent type, resulting in multiple target information sources.

[0198] In an exemplary embodiment, the retrieval module 706 is used to retrieve the target information source for each target information source based on intent representation and constraints, and determine the original text that matches the intent representation.

[0199] If the original text is of the unstructured text type, information extraction is performed on the original text to determine the information items corresponding to the intended representation.

[0200] In an exemplary embodiment, the information fusion module 708 includes a semantic alignment module, a confidence determination module, a conflict detection module, and a summary generative module. The semantic alignment module is used to perform semantic alignment on multiple information items to obtain the first candidate information item after semantic information alignment.

[0201] The confidence level determination module is used to determine the confidence level of each first candidate information item in each preset evaluation dimension according to a preset confidence evaluation model; and to determine the composite confidence level of the first candidate information item based on the confidence level.

[0202] The conflict detection module is used to perform conflict detection on each first candidate information item to obtain conflict data between the first candidate information items; and to generate corresponding conflict handling strategies based on the composite confidence level and the conflict data.

[0203] The summary generative module is used to determine multi-source information fusion prompts based on the first candidate information item, confidence level, composite confidence level, and conflict handling strategy, and to fuse multiple first candidate information items to generate a summary answer.

[0204] In an exemplary embodiment, the conflict detection module is configured to, if among the first candidate information items there are multiple second candidate information items with a composite confidence level greater than or equal to a first preset confidence level, generate a conflict handling strategy that includes sorting the multiple second candidate information items from largest to smallest according to their composite confidence levels, and determining the priority of each second candidate information item based on the sorting result.

[0205] And / or, if among the first candidate information items, there exists a third candidate information item with a composite confidence level greater than or equal to the second preset confidence level, and the conflicting data among multiple third candidate information items is non-conflicting, then the generated conflict handling strategy includes semantic fusion of multiple first candidate information items to generate a neutral summary; the second preset confidence level is greater than the first preset confidence level;

[0206] And / or, if there are multiple fourth candidate information items in the first candidate information items whose difference between any two composite confidence levels is within a preset range, and the conflicting data between the fourth candidate information items is conflicting, then the generated conflict handling strategy includes displaying the conclusion, information source, and composite confidence level of each fourth candidate information item.

[0207] In an exemplary embodiment, the multi-source information retrieval and fusion device described above further includes an information tracing module, which is used to determine the reference information of semantic segments in a summary answer. The reference information includes at least one of the following: information source, original text, and attribute information associated with the original text; generate an interactive reference of the semantic segment; and display the reference information referenced by the semantic segment in response to a trigger operation on the interactive reference.

[0208] In an exemplary embodiment, the information tracing module is used to split the summary answer according to a preset semantic format to obtain semantic fragments; for each semantic fragment, the semantic fragment and the original information fragment in the original text corresponding to the semantic fragment are encoded to obtain a first vector of the semantic fragment and a second vector of each original information fragment; based on the first vector and each second vector, the similarity between the semantic fragment and the original information fragment is determined, and a preset number of target original information fragments are determined from the original information fragments based on the similarity; the target original information fragments, the information source of the target original information fragments, and the attribute information associated with the target original information fragments are determined as the reference information of the semantic fragments in the summary answer.

[0209] Each module in the aforementioned multi-source information retrieval and fusion device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0210] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a multi-source information retrieval and fusion method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0211] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0212] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0213] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0214] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0215] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0216] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0217] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0218] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A multi-source information retrieval fusion method, characterized in that, The method comprises: analyzing the obtained user query data to determine a query intention type, an intention representation and a constraint condition of the query data; determining a plurality of target information sources matching the query intention type from a plurality of preset information sources; for each target information source, retrieving the target information source according to the intention representation and the constraint condition to obtain information items corresponding to the intention representation; performing fusion processing on the plurality of information items to generate a summarized answer; wherein the determining of the plurality of target information sources matching the query intention type from the plurality of preset information sources comprises: determining quality characteristic data of each information source in the plurality of information sources in different quality characteristic dimensions, and demand preference data of the query intention type for the quality characteristic dimensions; determining a matching degree matrix of the query intention type and the plurality of information sources according to the demand preference data and the quality characteristic data; determining, as the target information source matching the query intention type, an information source corresponding to an element value being a preset value in the matching degree matrix to obtain the plurality of target information sources.

2. The method of claim 1, wherein, The retrieving of the target information source according to the intention representation and the constraint condition to obtain the information items corresponding to the intention representation for each target information source comprises: for each target information source, retrieving the target information source according to the intention representation and the constraint condition to determine original text matching the intention representation; if the type of the original text is an unstructured text type, performing information extraction on the original text to determine the information items corresponding to the intention representation from the original text.

3. The method of claim 1, wherein, The fusion processing of the plurality of information items to generate the summarized answer comprises: performing semantic alignment on the plurality of information items to obtain first candidate information items after semantic information alignment corresponding to each information item; for each first candidate information item, determining a confidence degree of the first candidate information item in each preset evaluation dimension according to a preset confidence evaluation model; determining a composite confidence degree of the first candidate information item according to the confidence degree; performing conflict detection on each first candidate information item to obtain conflict data between the first candidate information items; generating a corresponding conflict processing strategy according to the composite confidence degree and the conflict data; determining a multi-source information fusion prompt according to the first candidate information item, the confidence degree, the composite confidence degree and the conflict processing strategy, fusing the plurality of first candidate information items to generate the summarized answer.

4. The method of claim 3, wherein, The generating of the corresponding conflict processing strategy according to the composite confidence degree and the conflict data comprises: if there are a plurality of second candidate information items with a composite confidence degree greater than or equal to a first preset confidence degree in the first candidate information items, the generated conflict processing strategy comprises sorting the plurality of second candidate information items according to the composite confidence degrees from large to small, and determining a priority of each second candidate information item based on a sorting result. And / or, if there are third candidate information items in the first candidate information items, the composite confidence of which is greater than or equal to a second preset confidence, and the conflict data between the plurality of third candidate information items is non-existent, the generated conflict processing strategy includes performing semantic fusion on the plurality of third candidate information items to generate a neutral summary; the second preset confidence is greater than the first preset confidence; And / or, if there are multiple groups of fourth candidate information items in the first candidate information items, any two of which have a difference in composite confidence within a preset range, and the conflict data between the fourth candidate information items is conflict, the generated conflict processing strategy includes displaying the conclusions, information sources, and composite confidence of each of the fourth candidate information items.

5. The method of claim 1, wherein, The method further comprises: determining reference information of a semantic segment in the summary answer, the reference information including at least any one of an information source, original text, and attribute information associated with the original text; generating an interactive reference of the semantic segment; in response to a trigger operation on the interactive reference, displaying the reference information referenced by the semantic segment.

6. The method of claim 5, wherein, The determination of the reference information of the semantic segment in the summary answer comprises: splitting the summary answer according to a preset semantic format to obtain semantic segments; encoding the semantic segments and original information segments in the original text corresponding to the semantic segments for each semantic segment to obtain a first vector of the semantic segment and a second vector of each original information segment; determining the similarity of the semantic segment and the original information segment according to the first vector and each second vector, and determining a preset number of target original information segments from the original information segments according to the similarity; determining the target original information segments, the information sources of the target original information segments, and the attribute information associated with the target original information segments as the reference information of the semantic segment in the summary answer.

7. A multi-source information retrieval fusion apparatus characterized by comprising: The device comprises: a data analysis module configured to analyze the obtained user query data, determine the query intent type, intent representation, and constraint conditions of the query data; an information source matching module configured to determine a plurality of target information sources matching the query intent type from a plurality of preset information sources; a retrieval module configured to retrieve each target information source according to the intent representation and the constraint conditions to obtain information items corresponding to the intent representation; an information fusion module configured to perform fusion processing on a plurality of information items to generate a summary answer; the information source matching module is further configured to determine quality characteristic data of each information source in different quality characteristic dimensions in the plurality of information sources, and demand preference data of the query intent type for the quality characteristic dimensions; determine a matching degree matrix of the query intent type and the plurality of information sources according to the demand preference data and the quality characteristic data; determine the information source corresponding to the preset value in the matching degree matrix as the target information source matching the query intent type to obtain a plurality of target information sources.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images