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

Abstract

The invention provides an information processing method and device, equipment, a medium and a program product, and relates to the field of financial science and technology or the field of big data. The method comprises the following steps: acquiring a knowledge document, sequentially carrying out blocking and segmentation processing on the knowledge document to obtain a plurality of to-be-processed text units, carrying out knowledge extraction on the to-be-processed text units to construct a knowledge graph and obtain query information, and retrieving a corresponding sub-graph from the knowledge graph based on the query information, according to the query information, contextual information corresponding to the query information is constructed, and the contextual information is input into a pre-trained knowledge model to obtain reply information corresponding to the query information; according to the method, high precision of information processing is ensured through fine processing, structured association, precise context construction and deep semantic understanding, so that the information processing accuracy is effectively improved.

Description

Technical Field

[0001] This application relates to the fields of financial technology or big data, and in particular to an information processing method, apparatus, equipment, medium and program product. Background Technology

[0002] In today's world, where the digital wave is sweeping the globe, data volume is exponentially exploding, and massive amounts of information are flooding in. Users not only face the dilemma of information overload and fragmentation, but also need to sift through information of varying quality to find valuable content. At the same time, users' needs for information processing have evolved from simple acquisition to precise semantic understanding, in-depth mining of cross-domain connections, and acquisition of structured answers.

[0003] Traditional information processing methods primarily rely on the keyword matching mechanism of traditional search engines. This mechanism breaks down user queries into independent keywords through word segmentation technology and uses inverted indexes to quickly locate web pages containing these keywords. Relevance ranking is usually calculated based on simple rules such as word frequency statistics and positional weights (e.g., titles, first paragraphs of the main text) to achieve information processing.

[0004] However, traditional search engines are limited by their keyword matching technology and have a relatively superficial understanding of the complex semantics, context, and user intent in natural language, resulting in low accuracy of information processing results. Summary of the Invention

[0005] This application provides an information processing method, apparatus, equipment, medium, and program product to solve the technical problem of low accuracy of information processing results caused by traditional information processing methods.

[0006] In a first aspect, this application provides an information processing method, including:

[0007] Collect knowledge documents and process them into blocks and segments in sequence to obtain multiple text units to be processed;

[0008] Knowledge is extracted from the text units to be processed in order to construct a knowledge graph;

[0009] Obtain query information and retrieve the corresponding subgraph from the knowledge graph based on the query information to construct the context information corresponding to the query information;

[0010] The context information is input into a pre-trained knowledge model to obtain the response information corresponding to the query information.

[0011] Secondly, this application provides an information processing apparatus, comprising:

[0012] The processing module is used to collect knowledge documents and sequentially divide the knowledge documents into blocks and segments to obtain multiple text units to be processed.

[0013] The first construction module is used to extract knowledge from the text units to be processed in order to construct a knowledge graph;

[0014] The second construction module is used to obtain query information and retrieve the corresponding subgraph from the knowledge graph based on the query information in order to construct the context information corresponding to the query information.

[0015] The response module is used to input the context information into a pre-trained knowledge model to obtain the response information corresponding to the query information.

[0016] Thirdly, embodiments of this application provide an information processing device, including: a memory and a processor;

[0017] The memory stores computer-executed instructions;

[0018] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0019] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0020] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0021] The information processing method provided in this application collects knowledge documents and sequentially segments and divides them into blocks to obtain multiple text units to be processed. Knowledge is extracted from these text units to construct a knowledge graph, query information is obtained, and corresponding subgraphs are retrieved from the knowledge graph based on the query information to construct context information corresponding to the query information. This context information is then input into a pre-trained knowledge model to obtain the response information corresponding to the query information. This method ensures high precision in information processing through refined processing, structured association, accurate context construction, and deep semantic understanding, thereby effectively improving the accuracy of information processing. Attached Figure Description

[0022] 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.

[0023] Figure 1 Flowchart of the information processing method provided in this application Figure 1 ;

[0024] Figure 2 Flowchart of the information processing method provided in this application Figure 2 ;

[0025] Figure 3 Flowchart of the information processing method provided in this application Figure 3 ;

[0026] Figure 4 A schematic diagram of the information processing apparatus provided in this application;

[0027] Figure 5 A schematic diagram of the information processing device provided in this application.

[0028] The accompanying drawings have illustrated specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0029] 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 apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0030] It should be noted that the information processing methods, apparatus, equipment, media and program products provided in this application can be used in the fields of fintech or big data, or in any other field. This application does not limit the application fields of the information processing methods, apparatus, equipment, media and program products.

[0031] In today's digital age, global data volume is experiencing exponential growth, with massive amounts of information overwhelming users. Faced with the dual challenges of information overload and fragmentation, users not only need to sift through this vast ocean of information for valuable content, but their needs are also evolving—from simple acquisition to the pursuit of precise semantic understanding, deep cross-domain connections, and structured answers. This evolution in demand places higher-dimensional requirements on information processing technologies.

[0032] Traditional search engines rely on keyword matching mechanisms, word segmentation technology, and inverted indexes to achieve fast retrieval, and depend on basic rules such as word frequency statistics and positional weights for relevance ranking. While this technical framework had an efficiency advantage in the early stages of information retrieval, its underlying logic remains at the level of mechanical matching of words at the surface level. When dealing with the complexity of natural language, this mechanism exposes fundamental flaws: it cannot parse the polysemy and metaphorical expressions of words, nor can it capture contextual relationships and deep semantics, and it lacks the ability to understand the user's true intentions.

[0033] To address the aforementioned issues, the information processing method provided in this application breaks down the massive and complex knowledge documents into multiple text units to be processed by segmenting and dividing the collected knowledge documents into blocks, thereby refining the granularity of knowledge. Next, knowledge is extracted from these text units and a knowledge graph is constructed to achieve structured association of knowledge. After obtaining the query information, the corresponding subgraph is retrieved based on the knowledge graph to construct contextual information. Finally, the contextual information is input into a pre-trained knowledge model, leveraging the model's semantic understanding and logical reasoning capabilities to extract the information's connotation and generate a response. This method ensures high precision in information processing through refined processing, structured association, accurate context construction, and deep semantic understanding, thereby effectively improving the accuracy of information processing.

[0034] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0035] Figure 1 Flowchart of the information processing method provided in this application Figure 1 In this embodiment, the executing entity is, for example, an information processing system. Figure 1 As shown, the method includes:

[0036] S101: Collect knowledge documents and process them into blocks and segments in sequence to obtain multiple text units to be processed.

[0037] Knowledge documents refer to the original text content obtained from data sources (such as books, papers, web pages, etc.).

[0038] Breaking down knowledge documents into smaller, structured text units transforms large, continuous texts into more manageable discrete units. For example, a long document might be broken down into paragraphs, sentences, or even smaller semantic units.

[0039] Segmenting and dividing knowledge documents into chunks improves the efficiency and accuracy of text processing. Directly processing large, continuous texts can lead to wasted computational resources or information redundancy. Breaking them down into smaller units allows for more refined analysis of each unit (such as named entity recognition and relation extraction). Furthermore, segmentation helps reduce the complexity of model input, avoiding performance degradation or information loss due to excessively long inputs, while also facilitating parallel processing and improving overall efficiency.

[0040] Specifically, knowledge documents are obtained through methods such as web scraping, APIs (Application Programming Interfaces), or file import. Based on the document structure (e.g., chapters, titles) or a fixed length (e.g., 500 words per group), the document is divided into multiple blocks, ensuring each block contains relatively independent content. Building upon this block division, the blocks are further subdivided into smaller units according to natural paragraphs or semantic boundaries (e.g., punctuation marks, line breaks), generating multiple text units to be processed.

[0041] Optionally, the knowledge document can be sequentially divided into blocks and segments to obtain multiple text units to be processed. Specific implementation methods include:

[0042] According to the document structure corresponding to the knowledge document, the knowledge document is divided into blocks; the text obtained after block division is segmented to obtain multiple text units to be processed.

[0043] The document is segmented according to its corresponding document structure. This means breaking down a long document into multiple blocks with independent themes or semantics based on its logical or physical structure (such as chapters, headings, paragraph separators, etc.). For example, a book is segmented by chapters, or a webpage is segmented by content blocks. Then, the text within each block is further segmented into smaller text units based on natural paragraphs or semantic boundaries (such as sentences, punctuation marks), ultimately generating multiple text units to be processed.

[0044] Knowledge documents have a hierarchical structure (such as chapters and sections). Segmenting them structurally preserves semantic integrity, avoids information fragmentation, and facilitates independent analysis of specific topic units. Directly processing large amounts of text can lead to wasted resources or noise interference. After segmentation, each unit is smaller, the model processes the data faster, and it can focus on key information. Segmented text units can be processed in parallel, significantly improving efficiency.

[0045] Specifically, the logical structure of the document is identified based on its format (such as PDF (Portable Document Format), HTML (HyperText Markup Language), or Markdown) or content characteristics (such as heading styles, indentation, and separators). The document is then divided into multiple blocks according to the parsed structure, each block containing a complete thematic unit (such as a chapter or subtopic). The text within each block is further segmented according to natural paragraphs (separated by line breaks or blank lines) or semantic boundaries (such as sentence endings), generating text units to be processed.

[0046] S102: Extract knowledge from the text units to be processed in order to construct a knowledge graph.

[0047] Among them, a knowledge graph is a semantic network composed of nodes (entities) and edges (relationships), which can intuitively display the relationships between entities and provide knowledge support for applications such as intelligent question answering and information processing.

[0048] Knowledge extraction and knowledge graph construction of text units involves extracting structured information such as entities (names, locations, organizations), relationships (belonging to, located in), and attributes (age, position) from unstructured or semi-structured text, and organizing this information in the form of a graph structure to form a knowledge graph.

[0049] Building knowledge graphs transforms scattered textual information into structured knowledge, thereby improving the understandability and usability of information. It integrates fragmented textual data into a unified knowledge representation, facilitating knowledge association across documents and domains. Structured knowledge graphs support relationship-based queries, rather than relying solely on keyword matching, improving retrieval accuracy.

[0050] S103: Obtain the query information and retrieve the corresponding subgraph from the knowledge graph based on the query information to construct the context information corresponding to the query information.

[0051] The query information refers to the query entered by the user (such as natural language questions or keywords).

[0052] Specifically, the user-input query (such as a natural language question or keywords) is parsed into structured information. Then, entities, relationships, and attributes related to the query are located within the constructed knowledge graph, and a local, structured subgraph is extracted. This subgraph contains only knowledge fragments directly related to the query, rather than the entire knowledge graph. Finally, the subgraph information is combined with the query itself to generate a context that includes the query intent, relevant knowledge, and background information.

[0053] Knowledge graphs are typically massive, making direct traversal inefficient. By retrieving subgraphs, the most relevant knowledge to the query can be quickly located, reducing interference from irrelevant information. Entities and relationships within the subgraph visually represent the semantic structure of the query, helping the system understand the query intent. Combining subgraphs with queries generates a context that includes both user needs and relevant background knowledge. Structured subgraph information facilitates rapid model processing, avoiding performance degradation caused by excessively long inputs.

[0054] S104: Input the context information into the pre-trained knowledge model to obtain the response information corresponding to the query information.

[0055] Specifically, inputting contextual information into a pre-trained knowledge model to generate a response means taking the previously constructed context (including query information, relevant subgraphs, and background knowledge) as input and passing it to a trained model (such as a large language model or knowledge reasoning model). The model then understands the semantics, entity relationships, and query intent within the context to generate a natural language response corresponding to the query. In other words, it transforms structured or semi-structured context into human-readable text answers, completing the closed loop from information retrieval to answer generation.

[0056] The information processing method provided in this embodiment collects knowledge documents and sequentially segments and divides them into blocks to obtain multiple text units to be processed. Knowledge is extracted from these text units to construct a knowledge graph, query information is obtained, and the corresponding subgraph is retrieved from the knowledge graph based on the query information to construct context information corresponding to the query information. The context information is then input into a pre-trained knowledge model to obtain the response information corresponding to the query information. This method ensures high precision in information processing through refined processing, structured association, accurate context construction, and deep semantic understanding, thereby effectively improving the accuracy of information processing.

[0057] Figure 2 Flowchart of the information processing method provided in this application Figure 2 .like Figure 2 As shown, in this embodiment... Figure 1 Based on the embodiments, the knowledge document is sequentially divided into blocks and segments to obtain multiple text units to be processed. This embodiment describes the information processing method, which includes:

[0058] S201: Determine whether a knowledge document contains hierarchical tags and obtain the corresponding judgment result.

[0059] By analyzing the format or content structure of knowledge documents, we can identify whether they contain tags with nested or hierarchical relationships (such as titles, chapters, sub-chaps, etc.). Hierarchical tags are usually manifested as heading-level divisions or explicit hierarchical markers within the document. The judgment result typically indicates whether the knowledge document has a text structure or not.

[0060] S202: When the judgment result indicates that the knowledge document has a text structure, the knowledge document shall be divided into blocks according to the document structure corresponding to the knowledge document.

[0061] When the judgment result indicates that the knowledge document has a text structure, it can be divided into blocks according to the specific structural features of the document (such as heading level, separators, paragraph indentation, etc.).

[0062] For example, for an HTML document, it can be based on <h1> arrive< / h1> <h6>The hierarchical relationship of tags divides each heading and its subordinate content into an independent block; for Markdown documents, the nesting relationship of heading symbols such as # and ## can be used to divide headings and their content into blocks at different levels; for PDF documents, text can be extracted through OCR and then divided into blocks according to the font size, bolding, or numbering of the headings (such as "1. Chapter" or "1.1 Sub-Chapter").

[0063] S203: When the judgment result indicates that the knowledge document does not have a text structure, the knowledge document is divided into blocks according to the preset text length.

[0064] When the judgment result shows that the knowledge document does not have a clear text structure (such as no hierarchical tags, no chapter division, or the content is continuous plain text without separation), the document can be evenly divided into blocks according to the pre-set text length threshold (such as every 500 characters, every 1000 words, or every 3 paragraphs).

[0065] Specifically, the document is divided into multiple sub-text blocks with a fixed number of characters, or split into blocks with a fixed number of paragraphs by counting the number of paragraphs, ensuring that the size of each block is controllable and the content is relatively complete. If the segmentation causes semantic breaks (such as truncating sentences), the boundaries can be adjusted or adjacent blocks can be merged to ultimately generate structured and easily processed text units.

[0066] S204: Perform segmentation on the text obtained after block division to obtain multiple text units to be processed.

[0067] Each text block is further subdivided based on natural paragraphs or semantic boundaries (such as line breaks, blank lines, punctuation marks, or sentence endings). For example, paragraphs separated by line breaks or blank lines are extracted independently from each block, or long paragraphs are divided into sentence-level units using punctuation marks (such as periods or question marks), ensuring that the generated text units are semantically complete and of appropriate size. Finally, each block is transformed into multiple text units to be processed (such as paragraphs or sentences).

[0068] Figure 3 Flowchart of the information processing method provided in this application Figure 3 .like Figure 3 As shown, in this embodiment... Figure 1 Based on the embodiments, the context information corresponding to the constructed query information is described in detail. The information processing method shown in this embodiment includes:

[0069] S301: Obtain the query information and extract the set of entities and relationships contained in the query information.

[0070] The core content is extracted from user-input text (such as natural language questions or search queries), for example, "Who is the founder of Company A?". Then, named entity recognition technology is used to extract a set of entities, such as "Company A" and "founder," and the semantic relationships between entities are further analyzed; for example, there is a "founding-creator" relationship between "Company A" and "founder." Finally, the entities and relationships are stored in a structured form (such as key-value pairs or graph nodes).

[0071] S302: Match at least one node that matches the entity set and at least one edge that matches the relation from the knowledge graph.

[0072] When matching entity sets and relationships from a knowledge graph, the entities extracted from the query information (such as "Company A" or "Founder") are compared with nodes in the graph. Nodes that are exact matches or semantically similar are found using node attributes (such as name and type) or unique identifiers (such as ID), for example, the node stored in the graph corresponds to "Company A". Simultaneously, the relationships in the query (such as "Founder") are matched with edges in the graph (the relationship types connecting nodes). For example, edges connecting nodes are found, or synonymous relationships are matched through semantic expansion (such as "Founder" corresponding to "B"). Finally, at least one node and at least one edge that meet the conditions are returned, forming a structured query result.

[0073] S303: Calculate correlation information based on the matched nodes and edges.

[0074] When calculating relevance information based on matched nodes and edges, it is necessary to comprehensively consider the degree of matching between the node and the query entity (such as name similarity and type consistency) and the relevance between the edge and the query relationship (such as direct matching or semantic synonym expansion). For example, if the query entity "Company A" matches a node in the knowledge graph, and the query relationship "Founder" corresponds to an edge in the graph, the relevance score can be quantified by the edge weight (such as the importance or frequency of the relationship), path length (such as the level of direct connection or indirect association), and the semantic relevance of the node attributes to the query context. Furthermore, if multiple matching nodes or edges exist, priority can be determined through weighted summation or ranking algorithms. Finally, the calculation results are transformed into quantifiable relevance indicators (such as scores or probabilities) to filter and rank candidate answers, ensuring that the returned results are highly relevant to the query intent.

[0075] S304: Construct a subgraph based on relevance information.

[0076] When constructing a subgraph based on relevance information, highly relevant nodes and edges should be used as core elements to filter out the knowledge fragments most closely related to the query intent. For example, based on relevance scores (such as nodes and edges above a threshold of 0.7), target nodes (such as "Company A" and its associated individuals) and relationship edges (such as "founder") can be extracted from the original knowledge graph, while retaining their direct or indirect context nodes (such as companies in the same industry and time attributes) to form a local, semantically complete subgraph structure.

[0077] S305: Based on the text units to be processed corresponding to the nodes contained in the subgraph, perform context assembly to obtain the context information corresponding to the query information.

[0078] When assembling context based on the text units to be processed corresponding to the nodes contained in the subgraph, the original text content (such as paragraphs, sentences, or entity descriptions) associated with each node in the subgraph needs to be reorganized according to semantic logic. For example, if a subgraph node corresponds to the introductory text of "Company A", the biographical description of the "founder", and the attribute value of "establishment time", then by identifying the referential relationships in the text (such as "this company" corresponding to "Company A") and temporal clues (such as the time association between "established in 1990" and "founder's resume"), the scattered text fragments can be pieced together into a coherent context, while retaining key relationships (such as "the founder joined in the year of establishment"). If there is redundant information, it can be simplified through summarization or keyword extraction. Finally, the assembled context information should completely cover the core content of the query requirements.

[0079] In one possible implementation, matching at least one node that matches the entity set and at least one edge that matches the relation from the knowledge graph further includes:

[0080] Expand at least one node that matches the entity set and at least one edge that matches the relation according to the pre-set number of hops;

[0081] Based on the matched nodes and edges, relevance information is calculated, including:

[0082] Based on the expanded nodes and edges, correlation information is calculated.

[0083] When matching entity sets and relationships from a knowledge graph, the first step is to compare the entities extracted from the query information (such as "Company A" or "Founder") with the nodes in the graph. Nodes that are exact matches or semantically similar are found using node attributes (such as name and type) or unique identifiers (such as ID), for example, the node in the graph corresponds to "Company A". Simultaneously, the relationships in the query (such as "Founder") are matched with the edges in the graph (the relationship types connecting nodes). For example, edges connecting nodes are found, or synonymous relationships are matched through semantic expansion (such as "Founder" corresponding to edges in the graph representing the same semantic meaning). Based on this, the matched nodes and edges are expanded according to a pre-set number of hops (such as 1 hop, 2 hops, or 3 hops). The number of hops limits the scope of the expansion; 1 hop represents directly connected nodes and edges, 2 hops include nodes and edges connected through an intermediate node, and so on. This expansion method allows for the extraction of information indirectly related to the query entities and relationships.

[0084] Based on the relevance information calculated from the expanded nodes and edges, it is necessary to analyze the degree of matching between the expanded nodes and the query entity. This involves not only direct name matching but also considering semantic similarity, type consistency, and other factors. For example, nodes matching the query entity "Company A," even if their names are not exactly the same, should be given a high matching score if they are highly similar in semantics such as industry or business scope. On the other hand, the fit between the expanded edges and the query relationship should be evaluated, including edge type weight, path length, and semantic relevance. For example, edges that directly match the query relationship have higher weights, while edges matched through semantic expansion have relatively lower weights. However, if the path is short and the semantic relevance is strong, their weights can be appropriately increased. Simultaneously, the overall position and influence of the expanded nodes and edges in the graph should be considered, such as whether the node is a core node or whether the edge connects multiple important nodes. A relevance score for each node and edge is obtained through quantitative calculations (such as weighted summation or machine learning model scoring).

[0085] In one possible implementation, after inputting contextual information into a pre-trained knowledge model to obtain the response information corresponding to the query information, the method further includes:

[0086] The response information is formatted according to a pre-set response framework to obtain and display response information that conforms to the framework.

[0087] After inputting contextual information into a pre-trained knowledge model to obtain the corresponding response to the query, the directly output response may suffer from issues such as loose format and inconsistent structure. For example, for some complex queries, the model's output response may contain a large amount of redundant information, or key information may be scattered throughout paragraphs, requiring users to spend considerable time extracting the core content. A pre-defined response framework, however, provides a clear structural template for the response, ensuring that the content is organized according to a specific format, highlighting key information, and improving readability and usability. Through formatting, the response can be transformed into a form that meets user expectations, facilitating quick understanding and use.

[0088] Pre-defined response frameworks are typically customized based on different query types and business needs. For example, for question-and-answer queries, the response framework might include a restatement of the question, the core answer, and supporting sources; for information summary queries, the framework might use lists, tables, or hierarchical structures to display the information. During formatting, the system will split, reorganize, and annotate the response information according to the requirements of the response framework. For example, key statements are extracted from the response information as the core answer and placed in the designated position within the framework; supporting information is organized and cited according to the framework's format, such as annotating data sources and quoting original text excerpts.

[0089] Formatted responses will be displayed according to a pre-defined response framework. The display method can be chosen based on the specific scenario, such as a clear page layout on a webpage, a simple card-style interface on mobile devices, or natural and fluent speech output using speech synthesis technology. On the one hand, users can quickly locate key information, improving information retrieval efficiency; on the other hand, the standardized format enhances the professionalism and credibility of the responses, improving the user experience.

[0090] Figure 4 A schematic diagram of the information processing apparatus provided in this application is shown below. Figure 4 As shown, the information processing device 400 provided in this embodiment includes:

[0091] The processing module 401 is used to collect knowledge documents and process them into blocks and segments in sequence to obtain multiple text units to be processed.

[0092] The first construction module 402 is used to extract knowledge from the text units to be processed in order to construct a knowledge graph;

[0093] The second construction module 403 is used to obtain query information and retrieve the corresponding subgraph from the knowledge graph based on the query information in order to construct the context information corresponding to the query information.

[0094] The response module 404 is used to input context information into a pre-trained knowledge model to obtain response information corresponding to the query information.

[0095] In one possible implementation, the information processing apparatus further includes: a block module 405;

[0096] The chunking module 405 is used to chunk the knowledge document according to the document structure corresponding to the knowledge document;

[0097] The processing module 401 is also used to segment the text obtained after block division to obtain multiple text units to be processed.

[0098] In one possible implementation, the information processing device further includes: a judgment module 406;

[0099] Module 406 is used to determine whether a knowledge document contains hierarchical tags and obtain the corresponding judgment result.

[0100] The knowledge document is sequentially divided into blocks and segments to obtain multiple text units to be processed, including:

[0101] The segmentation module 405 is also used to segment the knowledge document into blocks according to the document structure corresponding to the knowledge document when the judgment result indicates that the knowledge document has a text structure;

[0102] The segmentation module 405 is also used to segment the knowledge document into blocks according to a pre-set text length when the judgment result indicates that the knowledge document does not have a text structure.

[0103] The processing module 401 is also used to segment the text obtained after block division to obtain multiple text units to be processed.

[0104] In one possible implementation, the information processing device further includes: a matching module 407;

[0105] The processing module 401 is also used to obtain query information and extract the entity set and relationships contained in the query information;

[0106] Matching module 407 is used to match at least one node that matches the entity set and at least one edge that matches the relation from the knowledge graph;

[0107] The processing module 401 is also used to calculate correlation information based on the matched nodes and edges;

[0108] The first building module 402 is also used to build a subgraph based on correlation information;

[0109] The processing module 401 is also used to perform context assembly based on the text units to be processed corresponding to the nodes contained in the subgraph, so as to obtain the context information corresponding to the query information.

[0110] In one possible implementation, the information processing device further includes: an expansion module 408;

[0111] The extension module 408 is used to extend at least one node of the entity set and at least one edge of the relation according to a preset number of hops.

[0112] Based on the matched nodes and edges, relevance information is calculated, including:

[0113] Processing module 401 is used to calculate correlation information based on the expanded nodes and edges.

[0114] In one possible implementation, the information processing device further includes: a display module 409;

[0115] The display module 409 is used to format the response information according to a pre-set response framework in order to obtain and display response information that conforms to the response framework.

[0116] The information processing device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0117] Figure 5 A schematic diagram of the information processing device provided in this application. Figure 5 As shown, the electronic device of this embodiment may include: at least one processor 501; and a memory 502 communicatively connected to the at least one processor; wherein the memory 502 stores instructions executable by the at least one processor 501, the instructions being executed by the at least one processor 501 to cause the electronic device to perform the method as described in any of the above embodiments.

[0118] Optionally, the memory 502 can be either standalone or integrated with the processor 501. When the memory 502 is set up independently, the device also includes a bus for connecting the memory 502 and the processor 501.

[0119] The implementation principle and technical effects of the electronic device provided in this embodiment can be found in the foregoing embodiments, and will not be repeated here.

[0120] This application also provides a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed by a processor, the methods provided in any of the foregoing embodiments can be implemented.

[0121] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method provided in any of the foregoing embodiments.

[0122] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0123] It should be further noted that although the steps in the flowchart 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 flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0124] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0125] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0126] Unless otherwise specified, the processor can be any suitable hardware processor, such as a CPU, GPU, FPGA, DSP, and ASIC. Unless otherwise specified, the storage unit can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), enhanced dynamic random access memory (EDRAM), high-bandwidth memory (HBM), hybrid memory cube (HMC), etc.

[0127] If the integrated unit / module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0128] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0129] 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 following claims.

[0130] 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.< / h6>

Claims

1. An information processing method, characterized in that, include: Collect knowledge documents and process them into blocks and segments in sequence to obtain multiple text units to be processed; Knowledge is extracted from the text units to be processed in order to construct a knowledge graph; Obtain query information and retrieve the corresponding subgraph from the knowledge graph based on the query information to construct the context information corresponding to the query information; The context information is input into a pre-trained knowledge model to obtain the response information corresponding to the query information.

2. The method according to claim 1, characterized in that, The knowledge document is sequentially divided into blocks and segments to obtain multiple text units to be processed, including: The knowledge document is divided into blocks according to the document structure corresponding to the knowledge document; The text obtained after being segmented is further segmented to obtain multiple text units to be processed.

3. The method according to claim 1 or 2, characterized in that, Before sequentially performing block and segmentation processing on the knowledge document to obtain multiple text units to be processed, the method further includes: Determine whether the knowledge document contains hierarchical tags, and obtain the corresponding determination result; The knowledge document is sequentially divided into blocks and segments to obtain multiple text units to be processed, including: When the judgment result indicates that the knowledge document has a text structure, the knowledge document is divided into blocks according to the document structure corresponding to the knowledge document; When the judgment result indicates that the knowledge document does not have a text structure, the knowledge document is divided into blocks according to the preset text length; The text obtained after being segmented is further segmented to obtain multiple text units to be processed.

4. The method according to claim 1, characterized in that, The step of obtaining query information and retrieving the corresponding subgraph from the knowledge graph based on the query information to construct context information corresponding to the query information includes: Obtain the query information and extract the entity set and relationships contained in the query information; Match at least one node from the knowledge graph that matches the entity set and at least one edge from the relation that matches; Based on the matched nodes and edges, calculate the relevance information; Based on the correlation information, the subgraph is constructed; Based on the text units to be processed corresponding to the nodes contained in the subgraph, context assembly is performed to obtain the context information corresponding to the query information.

5. The method according to claim 4, characterized in that, The step of matching at least one node from the knowledge graph that matches the entity set and at least one edge that matches the relation further includes: According to a pre-set number of hops, at least one node of the entity set and at least one edge of the relationship are expanded. The calculation of relevance information based on the matched nodes and edges includes: The correlation information is calculated based on the expanded nodes and edges.

6. The method according to claim 1, characterized in that, After inputting the context information into the pre-trained knowledge model to obtain the response information corresponding to the query information, the method further includes: The response information is formatted according to a pre-set response framework to obtain and display response information that conforms to the framework.

7. An information processing device, characterized in that, include: The processing module is used to collect knowledge documents and sequentially divide the knowledge documents into blocks and segments to obtain multiple text units to be processed. The first construction module is used to extract knowledge from the text units to be processed in order to construct a knowledge graph; The second construction module is used to obtain query information and retrieve the corresponding subgraph from the knowledge graph based on the query information in order to construct the context information corresponding to the query information. The response module is used to input the context information into a pre-trained knowledge model to obtain the response information corresponding to the query information.

8. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-6.

Images