Intelligent service method and system based on deep reasoning and knowledge enhancement

Abstract

The invention is suitable for the technical field of intelligent questioning and answering, and particularly relates to an intelligent service method and system based on deep reasoning and knowledge enhancement, and the method comprises the steps: obtaining the identity information of a questioning student and target questioning information; the question and answer feature parameters are constructed by using the target question information, and the question understanding depth and precision are improved; screening candidate question and answer units from a preset question and answer knowledge base on the basis of the question and answer feature parameters, quickly positioning historical similar questions and corresponding answers, obtaining associated document fragments, and providing rich and authoritative knowledge sources for answer generation; when a candidate answer scheme is generated based on the candidate question and answer unit, content expansion and optimization are carried out by taking the candidate question and answer unit as a core carrier; according to student identity information, candidate answer schemes are screened by using a pre-trained deep reasoning screening model, the professionality and expression style of answers are dynamically adjusted, and a target answer scheme is displayed, so that pertinence and authority from question understanding, knowledge retrieval to answer generation are realized.

Description

Technical Field

[0001] This application belongs to the field of intelligent question answering technology, and in particular relates to an intelligent service method and system based on deep reasoning and knowledge enhancement. Background Art

[0002] With the in-depth application of artificial intelligence technology in education, office, and other fields, intelligent question-answering services based on large models have become a key technology for improving service efficiency and accuracy. Current mainstream intelligent question-answering services rely on preset rules or simple keyword matching to retrieve answers from a knowledge base when processing user questions. For example, when a user asks a question, the system simply analyzes the keywords in the question and searches the Q&A knowledge base for questions and corresponding answers that are similar in meaning.

[0003] However, existing intelligent question-answering services lack deep reasoning about users' underlying intentions, knowledge background, and behavioral patterns. When faced with semantically ambiguous questions or questions that integrate knowledge from multiple fields, they often fail to understand the essential meaning of the question and return irrelevant or even incorrect answers. For example, in a campus Q&A scenario, when a student asks about the specific regulations regarding failing a course at a certain college, existing question-answering models only match answers to superficial questions like "What does failing a course mean?" They fail to provide in-depth analysis and lack sufficient coverage of specialized knowledge in specific areas, resulting in answers that lack relevance and authority. Summary of the Invention

[0004] The embodiments of the present application provide an intelligent service method and system based on deep reasoning and knowledge enhancement, which can solve the problem that the answers provided by existing question-answering systems lack pertinence and authority.

[0005] In a first aspect, the embodiments of the present application provide an intelligent service method based on deep reasoning and knowledge enhancement, including: Obtain the identity information of the student asking the question and the target question information; Constructing question-answer feature parameters corresponding to the student asking the question according to the target question information; Based on the question and answer feature parameters, at least one candidate question and answer unit is selected from a preset question and answer knowledge base; wherein the candidate question and answer unit includes historical similar questions and corresponding answers, and associated document fragments; Generate at least one candidate answer solution based on the candidate question and answer unit; wherein any candidate answer solution uses any candidate question and answer unit as a core content carrier; The candidate answer solutions are screened using a preset deep reasoning screening model based on the identity information, a target answer solution is determined, and the target answer solution is presented to the student asking the question; wherein the deep reasoning screening model is a pre-trained machine learning model.

[0006] The above technical solutions in the embodiments of the present application have at least the following technical effects: The intelligent service method based on deep reasoning and knowledge enhancement provided by the embodiment of the present application accurately anchors user needs by obtaining the identity information of the student asking the question and the target question information. The target question information is used to construct question and answer feature parameters to improve the depth and accuracy of question understanding. Based on the question and answer feature parameters, candidate question and answer units are screened from the preset question and answer knowledge base, and similar historical questions and corresponding answers are quickly located, and related document fragments are obtained to provide a rich and authoritative source of knowledge for answer generation. When generating candidate answer solutions based on the candidate question and answer units, the content is expanded and optimized with the candidate question and answer units as the core carrier to ensure that the answer content has a solid foundation. According to the student's identity information, the candidate answer solutions are screened using a pre-trained deep reasoning screening model, and the professionalism, difficulty and expression style of the answer are dynamically adjusted and the target answer solution is displayed, so as to achieve pertinence and authority from question understanding, knowledge retrieval to answer generation.

[0007] In a second aspect, the embodiments of the present application provide an intelligent service system based on deep reasoning and knowledge enhancement, including: An acquisition unit, used to obtain the identity information of the student asking the question and target question information; A parameter unit, configured to construct question-answer feature parameters corresponding to the student asking the question based on the target question information; A first screening unit is configured to screen at least one candidate question and answer unit from a preset question and answer knowledge base based on the question and answer feature parameters; wherein the candidate question and answer unit includes historically similar questions and corresponding answers, and associated document fragments; A generating unit, configured to generate at least one candidate answer solution based on the candidate question and answer unit; wherein any candidate answer solution uses any candidate question and answer unit as a core content carrier; The second screening unit is used to screen the candidate answer solutions based on the identity information using a preset deep reasoning screening model, determine the target answer solution, and present the target answer solution to the student asking the question; wherein, the deep reasoning screening model is a pre-trained machine learning model.

[0008] In a third aspect, an embodiment of the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any one of the above aspects when executing the computer program.

[0009] In a fourth aspect, an embodiment of the present application provides a computer program product, which, when executed on an electronic device, enables the electronic device to execute the method according to any one of the above aspects.

[0010] It can be understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions of the above aspects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0012] Figure 1 This is a flowchart of an intelligent service method based on deep reasoning and knowledge enhancement provided by an embodiment of the present application; Figure 2 This is a schematic diagram of the structure of an intelligent service system based on deep reasoning and knowledge enhancement provided by an embodiment of the present application; Figure 3 It is a structural diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0013] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0014] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0015] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0016] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if the described condition or event is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of the described condition or event" or "in response to detecting the described condition or event," depending on the context.

[0017] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0018] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0019] Existing intelligent question-answering services lack deep reasoning about users' underlying intentions, knowledge background, and behavioral patterns. When faced with semantically ambiguous questions or questions that integrate knowledge from multiple fields, they often fail to understand the essential meaning of the question and return irrelevant or even incorrect answers. For example, in a campus Q&A scenario, when a student asks about the specific regulations regarding failing a course at a certain college, existing question-answering models only match answers to superficial questions like "What does failing a course mean?" They fail to provide in-depth analysis and lack in-depth coverage of specialized knowledge in specific areas, resulting in answers that lack pertinence and authority.

[0020] In order to solve the above problems, the embodiments of the present application provide an intelligent service method and system based on deep reasoning and knowledge enhancement. In this method, the user needs are accurately anchored by obtaining the identity information of the student asking the question and the target question information. The target question information is used to construct question and answer feature parameters to improve the depth and accuracy of question understanding. Based on the question and answer feature parameters, candidate question and answer units are screened from the preset question and answer knowledge base, and historical similar questions and corresponding answers are quickly located, and related document fragments are obtained to provide a rich and authoritative source of knowledge for answer generation. When generating candidate answer solutions based on candidate question and answer units, the content is expanded and optimized with the candidate question and answer units as the core carrier to ensure that the answer content has a solid foundation. According to the student identity information, the candidate answer solutions are screened using a pre-trained deep reasoning screening model, and the professionalism, difficulty and expression style of the answer are dynamically adjusted and the target answer solution is displayed, so as to achieve pertinence and authority from question understanding, knowledge retrieval to answer generation.

[0021] The intelligent service method based on deep reasoning and knowledge enhancement provided in the embodiment of the present application can be applied to electronic devices. In this case, the electronic device is the executor of the intelligent service method based on deep reasoning and knowledge enhancement provided in the embodiment of the present application. The embodiment of the present application does not impose any restrictions on the specific type of electronic device.

[0022] For example, the electronic device may be a mobile phone, tablet computer, wearable device, in-vehicle device, augmented reality (AR) / virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), desktop computer, smart screen, smart TV and other terminal devices.

[0023] In order to better understand the intelligent service method based on deep reasoning and knowledge enhancement provided in the embodiment of the present application, the specific implementation process of the intelligent service method based on deep reasoning and knowledge enhancement provided in the embodiment of the present application is exemplarily introduced below.

[0024] Figure 1 A schematic flow chart of an intelligent service method based on deep reasoning and knowledge enhancement provided by an embodiment of the present application is shown. The intelligent service method based on deep reasoning and knowledge enhancement includes: S100, obtaining the identity information of the student asking the question and the target question information.

[0025] It can be understood that the identity information of the student asking the question is a basic identity identifier, which may include implicit characteristic information such as the student's historical learning records and knowledge mastery. Identity information can be obtained through active user input, system account authentication, and campus information platform docking. For example, when students log in to the learning system, the identity data associated with the account is automatically synchronized, or personal information is submitted by filling out a form. The target question information refers to the specific content of the question asked by the student, including the question text, text converted from voice input, and accompanying pictures or documents and other information carriers. When obtaining the target question information, the question text can be directly collected through the text input box, the voice question can be converted into text through automatic speech recognition (ASR), and the questions in the picture can be extracted through optical character recognition (OCR). At the same time, the format of the original information obtained is standardized, such as removing special characters and unifying text encoding to ensure the integrity and processability of the information, and provide accurate basic data for subsequent question and answer processing.

[0026] In some embodiments, before obtaining the identity information of the student asking the question and the target question information, the method further includes: S600, obtaining documents related to question and answer content, and building a question and answer knowledge base based on the documents related to question and answer content; wherein the question and answer knowledge base includes a first knowledge warehouse built based on domain labels, and a second knowledge warehouse built based on semantic feature vectors.

[0027] It can be understood that building a question-and-answer knowledge base is a process of structuring unstructured documents. Relevant documents such as textbooks, papers, and online questions and answers can be obtained through web crawlers, document import, manual collection, etc., and seed question-and-answer units with labeled domain labels can be extracted. Seed question-and-answer units refer to the most basic and core question-and-answer combination that constitutes the question-and-answer system. For example, regular expressions can be used to match the "question:... answer:..." format, or information extraction models (such as BERT sequence labeling) can be used to identify question-and-answer boundaries. Semantic analysis is performed on the seed units, and feature vectors are generated using models such as SBERT to build a semantic index database (such as Milvus, which maps storage vectors to question-and-answer units). For unlabeled question-and-answer units, after generating feature vectors, the similarity is calculated with the vectors in the index library. The most similar seed unit is found through the KNN algorithm (K=5), and the domain label with the majority vote is taken to complete automatic labeling. Finally, the seed question-and-answer units are clustered according to the domain label to form the first knowledge warehouse, and the semantic feature vectors are stored to form the second knowledge warehouse, thus building a composite knowledge base that supports domain classification and semantic retrieval. The dual-warehouse design of building the first knowledge warehouse based on domain labels and the second knowledge warehouse based on semantic feature vectors effectively prevents the missing retrieval dimensions and complex problem matching deviations under a single knowledge management model (preventing cross-domain problems from mismatching unrelated knowledge in a single domain and affecting authority). By limiting the domain scope through structured classification of the label warehouse and supplementing the semantic association through vector similarity of the semantic warehouse, a two-way verification architecture of "classification prevention of transgression and semantics prevention of omission" is formed, ensuring the comprehensiveness of knowledge retrieval, the stability of updates and the accuracy of matching, and providing intelligent question and answer with a robust knowledge management solution that combines rule certainty and data flexibility.

[0028] In one possible implementation, S600, obtaining documents related to question and answer content and building a question and answer knowledge base based on the documents related to question and answer content, includes: S610: Obtain documents related to the question and answer content, and extract seed question and answer units marked with domain labels from the documents related to the question and answer content.

[0029] It can be understood that when obtaining documents related to Q&A content, channels such as the school's internal OA system, the teaching affairs office platform, and the college official website can be used to collect school internal regulation documents such as the "Student Handbook", the "Teaching Management Regulations", and the "Laboratory Safety Norms", as well as structured Q&A data such as counselor Q&A records and common questions FAQ in the teaching affairs system. The seed Q&A unit is a basic Q&A data unit that is authoritative, typical, and can be used to derive associated Q&As. For example, from the document "XX University Student Disciplinary Action Measures", the typical Q&A format of "Question: How to deal with cheating in exams? Answer: According to Article X of the school regulations, a demerit will be given and the qualification for applying for a degree will be cancelled" is matched through regular expressions, or the boundaries of Q&As such as "Question: What are the punishment measures for coming back to the dormitory late? Answer: Returning to the dormitory after 23:00 is considered coming back late, and being reported for criticism will be given if the number of times accumulates more than 3 times" are identified using syntactic analysis. Domain label annotation can be combined with the hierarchical structure of the school internal regulation system. When manually annotating, the domain can be labeled according to the document chapter attribution. For example, the Q&As in the "Student Status Management Regulations" are labeled as "School Internal Regulations - Teaching Management - Student Status Management"; when using semi-supervised learning for annotation, the preset label dictionary can be matched through the keywords in the document title (such as associating the label "School Internal Regulations - Life Management - Dormitory Management" if the title contains "dormitory"), or the domain label can be automatically inherited using the school internal regulation classification tree (such as "School Internal Regulations → Teaching Management → Course Assessment → Cheating Handling"). The extracted seed Q&A units need to be associated with the specific articles of the school internal regulations. For example, "Question: How to apply for course make-up? Answer: An application needs to be submitted through the teaching affairs system before the 4th week of each semester. See Article 5 of the 'XX University Course Make-up Management Measures' Label: School Internal Regulations - Teaching Management - Course Make-up", so that the Q&A content can be traced back to the specific school regulation basis, providing an authoritative answer basis for subsequent student questions.

[0030] S620, perform semantic analysis on each seed Q&A unit to generate a feature vector corresponding to the seed Q&A unit.

[0031] It can be understood that semantic analysis is a process of converting text into numerical feature vectors. Perform preprocessing such as word segmentation, stop word removal (such as "de", "le"), and word form reduction (such as "running" → "run") on the question and answer texts of the seed Q&A units to ensure the consistency of lexical representation. The feature vectors generated by semantic analysis can represent the semantic essence of the Q&A units, providing a numerical basis for subsequent semantic retrieval and similarity calculation.

[0032] S630, construct a semantic index database based on the feature vectors, and use the semantic index database to automatically assign domain labels to Q&A units without labeled domain labels.

[0033] It can be understood that when constructing a semantic index database based on feature vectors, the feature vectors of seed question-and-answer units such as campus regulations and campus activities (such as 768-dimensional vectors generated by BERT) can be stored in a vector database such as Milvus or Faiss, establishing a mapping relationship between "feature vector-question-and-answer unit-labeled domain label" to form an index structure that supports efficient semantic retrieval. When processing question-and-answer units without labeled domain labels, the target feature vector is first generated through the same semantic analysis process. Then, K candidate seed question-and-answer units (K=5) whose cosine similarity with this vector exceeds a threshold (such as 0.7) are retrieved from the index database. The textual semantic consistency between the candidate question-and-answer unit's domain label and the current question-and-answer unit is calculated (such as using Sentence-BERT to calculate question text similarity). The seed question-and-answer unit with the highest matching domain label (such as "campus regulations-teaching management-scholarship management") is selected and assigned to the unlabeled question-and-answer unit, achieving automated domain classification of campus regulations questions and answers, ensuring that new questions and answers are accurately mapped to the corresponding categories of the campus regulations system.

[0034] Optionally, in step S630, automatically assigning domain labels to question-answer units that are not labeled with domain labels using a semantic index database includes: S631: Perform semantic analysis on the question-answer unit without domain labels to generate a target feature vector.

[0035] It can be understood that for unlabeled school regulations question and answer units (such as "Q: What procedures are required to resume classes after a leave of absence?"), text preprocessing can be performed first to remove non-critical information such as "processing" and "completed", and standardize terms such as "leave of absence and resumption of classes" and "procedures". Then, by extracting semantic features, the question and answer text can be converted into a 768-dimensional target feature vector. The target feature vector contains semantic representations of keywords such as "leave of absence", "resumption of classes", and "procedures", as well as implicit relationships such as "process" and "school rules and regulations", providing a numerical basis for subsequent vector similarity calculations with seed question and answer units in the index library.

[0036] S632: Calculate the similarity between the target feature vector and each feature vector in the semantic index database, and select candidate seed question-answer units whose similarity exceeds a threshold.

[0037] It can be understood that the ANN retrieval function of the Milvus vector database can be used to calculate the cosine similarity between the target feature vector of the unlabeled question and answer unit and the feature vectors of all seed question and answer units in the index library, and set a threshold (such as 0.7) to screen out candidate seed units with a similarity higher than the threshold. For example, 5 seed questions and answers with a similarity > 0.7 to the question vector of "leave of absence and resumption of classes" were retrieved from the campus regulations index library, including "Q: How to apply for leave of absence?", "Q: What materials need to be submitted for resumption of classes?", etc., to ensure that the candidate question and answer units are highly relevant to the current question at the semantic level.

[0038] S633: Perform semantic consistency check on the candidate seed question and answer units to determine the target seed question and answer unit with the highest matching degree.

[0039] It can be understood that for the screened candidate seed question and answer units, in addition to vector similarity, the semantic consistency of the question text is further calculated through text matching algorithms (such as edit distance, Jaccard coefficient) and semantic models (such as Sentence-BERT). For example, the text similarity between the unlabeled question "What procedures are required to resume classes after a leave of absence?" and the candidate question and answer seed unit "Q: What is the process for students on leave to resume classes?" is 0.85, which is higher than other candidate question and answer units. Both involve core semantics such as "leave of absence and resumption of classes" and "process procedures". This candidate question and answer unit is determined to be the target seed unit with the highest matching degree, avoiding labeling errors caused by high vector similarity but actual semantic ambiguity (such as the vectors of "leave of absence" and "drop out" are close but the fields are different).

[0040] S634: Assign the domain label of the target seed question-answering unit to the question-answering unit.

[0041] It can be understood that the domain label of the target seed unit with the highest matching degree (such as "campus regulations-life management-student status changes") is directly assigned to the unlabeled question and answer unit to complete the domain classification. For example, the above-mentioned "leave of absence and resumption of classes procedures" question is assigned the same label as the target seed unit, and the source of the school regulations corresponding to the label is recorded (such as Chapter 7 of "XX University Student Status Management Regulations") to ensure that the labeling results can be traced back to the specific category of the school regulations system, providing accurate classification for subsequent question and answer matching based on domain labels.

[0042] S640: Cluster all seed question-answer units based on domain labels to form a first knowledge warehouse.

[0043] It can be understood that according to the hierarchical system of school regulations (such as "school regulations → teaching management → course assessment → cheating handling"), all seed question and answer units are hierarchically clustered according to field labels, and the K-means algorithm or the label tree-based clustering method is used to classify question and answer units with the same or similar field labels into the same category. For example, questions and answers such as "scholarship application" and "credit grade point calculation" are classified into the "school regulations-teaching management-academic awards" category, forming a first knowledge warehouse with a hierarchical structure. Each category corresponds to a sub-field of the school regulations, supporting rapid retrieval of related question and answer units through field labels, thereby improving the retrieval efficiency and organization of the knowledge base.

[0044] S200, constructing question and answer feature parameters corresponding to the student asking the question based on the target question information.

[0045] It can be understood that constructing question-answering feature parameters involves converting question text into structured semantic features using natural language processing techniques. The target question information can be segmented, tagged with parts of speech, and subjected to dependency parsing to extract keywords (e.g., "deep learning," "model," and "training" in "deep learning model training") and entity relationships, forming a semantic knowledge subgraph (with nodes representing entities and edges representing relationships). The question is then encoded to generate a high-dimensional feature vector (e.g., 768 dimensions) that incorporates contextual semantics. This high-dimensional feature vector can capture the implicit semantics of the question (e.g., the relationship between "optimizing training efficiency" and "learning rate adjustment strategy"). Keyword matching can be used with domain dictionaries or text classification models (e.g., TextCNN) to determine the domain label of the question (e.g., "computer science - machine learning") and extract semantic index fragments (e.g., the core term combination in the question). The generated question-answering feature parameters, including the domain label, semantic feature vector, and knowledge subgraph structure, comprehensively represent the semantic essence of the question and provide a basis for accurate matching in knowledge base retrieval.

[0046] In one possible implementation, S200 constructs question-answer feature parameters corresponding to the student asking the question based on the target question information, including: S210 , performing semantic analysis according to the target question information to determine a semantic knowledge subgraph corresponding to the target question information and a semantic index segment of the semantic knowledge subgraph.

[0047] It can be understood that when semantically parsing the target question information, key entities (such as "exam cheating" and "return to graduate school qualification") and predicate relationships (such as "cancel" and "affect") can be identified through word segmentation and part-of-speech tagging, and a semantic knowledge subgraph can be constructed. This is done with entities as nodes and relationships as edges. For example, the "exam cheating" node is connected to the "return to graduate school qualification cancellation" node via the "result" edge. Simultaneously, semantic index fragments are extracted, which are the core term combinations in the question (such as "exam cheating" and "return to graduate school qualification cancellation"). These core term combination fragments serve as keywords for searching the knowledge base and are also used in the subsequent construction of semantic structure features, ensuring that the knowledge subgraph intuitively reflects the semantic network and core retrieval elements of the question.

[0048] S220, constructing a semantic adjacency tensor corresponding to the semantic knowledge subgraph based on the semantic index fragment.

[0049] It can be understood that when converting a semantic knowledge subgraph into a semantic adjacency tensor, the entities in the semantic index fragment are used as dimensions to construct a multidimensional tensor representing the strength of associations between nodes. For example, for the question "Does cheating on an exam affect admission to graduate school?", the corresponding position value of the "cheating on an exam" and "graduate school admission qualification" nodes in the tensor is 0.9 (based on co-occurrence frequency and syntactic relationship weights). The semantic adjacency tensor reflects the topological structure of the knowledge subgraph, numerically representing the closeness of semantic associations, indicating that it can efficiently store semantic relationships and providing a mathematical foundation for subsequent topological sorting.

[0050] S230, topologically sorting the semantic adjacency tensor to generate a semantic subgraph dependency tree; wherein the semantic subgraph dependency tree explicitly expresses the hierarchical dependency relationship between semantic knowledge subgraphs in a tree structure.

[0051] As you can understand, when topologically sorting the semantic adjacency tensor, the logical order is determined based on the association weights between nodes, generating a tree-structured semantic subgraph dependency tree. For example, "exam cheating" is the root node, "school regulations and sanctions," "return to graduate school policy" are first-level child nodes, and "disqualification" is the second-level child node. The edge weights represent the strength of the dependency (e.g., "exam cheating → school regulations and sanctions" has a weight of 0.8). The semantic subgraph dependency tree clearly displays the hierarchical relationships (e.g., causality, subordination) between the semantic elements in the question, transforming the abstract semantic network into an intuitive logical hierarchy, facilitating subsequent hierarchical feature extraction and structured answer generation.

[0052] S240, performing hierarchical decomposition on the semantic subgraph dependency tree to determine question-answer feature parameters corresponding to the student asking the question.

[0053] As can be understood, when performing a hierarchical decomposition of the semantic subgraph dependency tree, core features are extracted according to the tree's hierarchical structure (root node → child node → leaf node): the root node corresponds to the domain label of the question (e.g., "campus regulations - teaching management - examination discipline"), the child nodes correspond to detailed knowledge points (e.g., "cheating penalties," "requirements for admission to graduate school"), and the edge weights correspond to feature importance. Simultaneously, combined with the semantic feature vectors generated by a pre-trained language model (such as BERT), the final result is a question-answering feature parameter consisting of the domain label, the semantic feature vector, and the hierarchical weights. For example, the decomposition yields the domain label "campus regulations," a semantic vector (encoding semantics such as "exam cheating" and "requirements for admission to graduate school"), and hierarchical features (weight vectors for "cheating penalties → impact on admission to graduate school"). This comprehensively captures the semantic essence and logical hierarchy of the question, transforming semantics from text strings into computable semantic links through the semantic subgraph dependency tree. This approach is particularly applicable to domains such as school regulations and laws, where precise, logical, and authoritative matching is required.

[0054] S300, screening out at least one candidate question and answer unit from a preset question and answer knowledge base based on question and answer feature parameters; wherein the candidate question and answer unit includes historical similar questions and corresponding answers, and associated document fragments.

[0055] It can be understood that the question and answer feature parameters include domain labels and semantic feature vectors. Based on the domain labels, the question and answer unit set in the same field can be quickly located in the first knowledge warehouse through a tree index (such as all FAQs under the category of "campus regulations-teaching management-examination discipline"), and then the semantic feature vector can be used to retrieve semantically similar question and answer units in the vector database of the second knowledge warehouse (using cosine similarity calculation to match the semantic association between "Is cutting the screen in online exams considered cheating" and the historical question "Zoom exam window switching judgment rules"). Then, the matching coefficients of each question and answer unit in the two sets are calculated separately, namely the first matching coefficient and the second matching coefficient. Finally, through cross-fusion, historical similar questions and answers that are related to the field and semantically similar, and related document fragments are screened out by sorting by scores to form a candidate question and answer unit set, providing highly relevant knowledge materials for the subsequent generation of answer solutions.

[0056] In one possible implementation, the question-answer feature parameters include a domain label and a semantic feature vector. S300 , screening at least one candidate question-answer unit from a preset question-answer knowledge base based on the question-answer feature parameters, includes: S310, matching the corresponding first question and answer unit set from the first knowledge warehouse according to the domain label, and retrieving the corresponding second question and answer unit set from the second knowledge warehouse according to the semantic feature vector.

[0057] It can be understood that candidate question and answer units can be simultaneously retrieved from the first and second knowledge repositories based on the domain label and semantic feature vector in the question and answer feature parameters. Searches in the first knowledge repositories can use the domain label as the index key to directly locate the set of question and answer units in the corresponding category through a tree-like hierarchical structure (e.g., "campus regulations → teaching management → scholarship management"). For example, when the domain label of a question is "scholarship application conditions," all FAQs related to the "Scholarship Management Measures" can be quickly retrieved from the first knowledge repositories, ensuring strict alignment between the search content and the question topic. Searches in the second knowledge repositories (semantic vector repositories) can input the semantic feature vector of the question (e.g., a 768-dimensional vector generated by BERT) into the vector database and, through cosine similarity calculation, retrieve the historical question and answer units with the closest semantics. For example, the semantic vector of the question "Can I apply for a national scholarship with a GPA of 3.2?" can match the historical question and answer "Can I apply with a GPA of 3.5 or above?" Even if the keywords "GPA" and "GPA" are expressed differently, generalized search can be achieved through semantic association.

[0058] S320: Determine a first matching coefficient list for the first question and answer unit set and a second matching coefficient list for the second question and answer unit set respectively.

[0059] It can be understood that when determining the first matching coefficient list, for each unit in the first question and answer unit set, the coefficient is calculated from three aspects: domain label matching, historical adoption rate, and historical follow-up rate. The domain label matching is determined by the similarity of the label tree hierarchy (for example, when the target label is level four and the candidate label is level three, the matching is level three divided by level four), the historical adoption rate is the number of times the question and answer is adopted by the user divided by the total number of displays, and the historical follow-up rate is the number of times the user continues to ask questions after adopting the question and answer divided by the number of adoptions. The final first matching coefficient is the weighted sum of the domain label matching, the historical adoption rate, and the historical follow-up rate. When determining the second matching coefficient list, for each unit in the second question and answer unit set, the weighted calculation is based on the cosine similarity of the semantic vector and the question and answer completeness. Cosine similarity reflects the degree of semantic similarity between the question and the candidate question and answer, and the question and answer completeness is the proportion of the answer that contains the core elements of the question. The second matching coefficient is the cosine similarity multiplied by the question and answer completeness, thereby forming a numerical list that quantifies the matching degree of each question and answer unit, providing an accurate quantitative basis for subsequent cross-fusion.

[0060] Optionally, S320, respectively determining a first matching coefficient list for the first question-and-answer unit set and a second matching coefficient list for the second question-and-answer unit set includes: S321, calculate the degree of matching between each first question and answer unit in the first question and answer unit set and the target question information, obtain the first matching coefficient of each first question and answer unit, and form all the first matching coefficients into a first matching coefficient list; wherein, the first matching coefficient is calculated through the domain label, the historical adoption rate score, and the historical follow-up rate.

[0061] It is understandable that the degree of matching for each first question-and-answer unit can be quantified from multiple dimensions. For example, first extract its domain label and conduct semantic analysis with the domain label of the target question (such as calculating label path similarity based on WordNet or counting overlap by co-occurrence frequency), then calculate the adoption rate of the unit in historical interactions (the ratio of the number of times it is confirmed to be valid by users to the number of times it is displayed), and at the same time convert the frequency of subsequent user questions into an inverse score (such as 1-question rate to reflect the adequacy of the answer), and finally, weighted sum the three indicators according to preset weights (for example, 40% for domain label, 35% for historical adoption rate, and 25% for historical question rate) to obtain the first matching coefficient of the first question-and-answer unit.

[0062] S322, calculate the degree of matching between each second question and answer unit in the second question and answer unit set and the target question information, obtain the first matching coefficient of each second question and answer unit, and form all the second matching coefficients into a second matching coefficient list; wherein, the second matching coefficient is obtained by calculating the cosine similarity and the question and answer completeness.

[0063] It's understandable that each second question-and-answer unit can be evaluated based on semantic relevance and content completeness. Cosine similarity can be calculated between its semantic feature vector (e.g., generated through BERT model encoding) and the semantic vector of the target question to quantify the degree of semantic match between the texts. Information entropy and keyword coverage are then used to assess the answer's coverage of the core elements of the question (e.g., whether it includes key information such as arguments, steps, and conclusions). Finally, these two metrics are weighted and combined according to preset weights (e.g., 60% for cosine similarity and 40% for question-and-answer completeness) to obtain the second matching coefficient for the unit.

[0064] S330, cross-merging the first question and answer unit set and the second question and answer unit set based on the first matching coefficient list and the second matching coefficient list to obtain a cross-question and answer unit set; wherein the cross-question and answer unit set includes at least one cross-question and answer unit.

[0065] It can be understood that dual-set fusion can be achieved through multi-dimensional association calculation, and the question and answer units in the two sets are combined by Cartesian product to generate a candidate set of unit pairs. Then, for each unit pair, the three-dimensional association is calculated, including domain label matching (such as semantic path similarity based on WordNet), semantic feature cosine similarity (such as calculated by BERT vector), and the product of the first matching coefficient and the second matching coefficient. Then, a comprehensive association score is generated according to the preset weights (such as 30% for domain matching, 40% for semantic similarity, and 30% for coefficient product). Finally, through double threshold screening (such as filtering low-association pairs with the basic threshold, and then retaining the top k% according to the score) and combined with the maximum marginal correlation algorithm (MMR), semantic redundant unit pairs are eliminated, and finally a cross-question and answer unit set containing high-value combinations is formed.

[0066] S340: Determine each cross-question and answer unit in the cross-question and answer unit set as a candidate question and answer unit.

[0067] It can be understood that all cross-question and answer units in the set can be traversed, and each cross-question and answer unit can be directly marked as a candidate state, providing a basic data unit for the screening, evaluation or application of subsequent candidate question and answer units, ensuring that each valid unit in the set is qualified as a candidate, and laying the foundation for subsequent processing procedures.

[0068] S400, generating at least one candidate answer solution based on the candidate question and answer unit; wherein any candidate answer solution uses any candidate question and answer unit as a core content carrier.

[0069] It can be understood that the answer plan can be constructed based on the candidate question and answer units, that is, for each candidate question and answer unit, a complete answer framework is generated by integrating relevant semantic information to ensure that the answer plan closely revolves around the core of the question. At the same time, a multi-candidate mechanism is used to provide diverse answer possibilities, providing sufficient options for subsequent solution screening or optimization.

[0070] In one possible implementation, S400, generating at least one candidate answer solution based on the candidate question-answer unit, includes: S410, for each candidate question and answer unit, calculate the correlation between the candidate question and answer unit and the hierarchical nodes of the semantic subgraph dependency tree, and extract the node document fragments corresponding to the hierarchical nodes with a correlation greater than a threshold.

[0071] It can be understood that the cosine similarity calculation can be performed between the semantic feature vector of the candidate question and answer unit and the feature vector of each hierarchical node in the semantic subgraph dependency tree (including semantic representations such as domain concepts and relationship edges), or by calculating the attention weights between the nodes, the degree of semantic association between the two can be quantified. Then, by pre-setting a correlation threshold (such as 0.6), the hierarchical nodes with a correlation exceeding the threshold are screened out, and the document fragments corresponding to these hierarchical nodes in the knowledge graph are extracted (such as text blocks containing concept definitions, formula derivations, case descriptions, etc.).

[0072] S420: Expand and optimize the candidate question-answer units through natural language generation based on the extracted node document fragments to generate candidate answer solutions.

[0073] It can be understood that the original answer content of the candidate question-answering unit and the extracted node document fragment can be used as input, and the key information in the answer and the supplementary knowledge of the document fragment can be aligned through the attention mechanism. Then, the content is reorganized according to the preset logical structure (such as question analysis-principle explanation-case evidence). At the same time, the professional terms and key data in the document fragment are retained through the copy mechanism. Finally, a fluent and natural text is generated through the decoder to form a candidate answer solution that contains in-depth background information, extended argumentation or example support, thereby improving the richness and professionalism of the answer.

[0074] S500, using a preset deep reasoning screening model to screen candidate answer solutions based on identity information, determine a target answer solution, and present the target answer solution to the student asking the question; wherein the deep reasoning screening model is a pre-trained machine learning model.

[0075] It is understandable that the "one-size-fits-all" model of traditional question-answering systems can be broken through the complete chain from user feature analysis to accurate answer recommendation through deep reasoning technology. Different students have significant differences in knowledge reserves, learning habits, professional needs, etc. For example, computer science students have a much deeper understanding of algorithmic problems than non-majors, while liberal arts students pay more attention to the logic and expression style of the answers. The deep reasoning screening model can simulate the human cognitive process by integrating the multi-dimensional features of student identity information, and filter out the answer that best meets the user's needs from the set of candidate answer solutions. The deep reasoning screening model can be trained based on historical interaction data. Through architectures such as Transformer and graph neural networks, it learns the complex mapping relationship between user features and answer quality, thereby realizing intelligent and personalized screening of candidate answer solutions, ensuring that the answers displayed to users meet both knowledge accuracy and their personalized needs.

[0076] In one possible implementation, S500 uses a preset deep reasoning screening model to screen candidate answer solutions based on the identity information, determines a target answer solution, and presents the target answer solution to the student asking the question, including: S510, extracting the student's professional background feature vector and historical question and answer preference feature vector based on the identity information.

[0077] It's understandable that representative features can be extracted from identity information. Student identity information typically includes multi-source, heterogeneous data, including basic attributes (such as grade and major), learning records (course grades, electives), and interaction logs (historical questions and follow-up questions). When extracting a professional background feature vector, structured data such as the major name and major courses can be encoded. For example, "Computer Science and Technology" can be mapped into a specific embedding vector. This, combined with the prerequisite relationships and knowledge point associations between courses in the professional knowledge graph, can construct a feature vector that captures the professional knowledge structure. For historical question-answering preference feature vectors, natural language processing techniques can be used to analyze the semantic content, question type (factual, inferential), and follow-up question patterns (questioning answers, seeking further explanations) of historical questions. Recurrent neural networks (RNNs) or Transformer models can be used to capture long-term dependencies in question sequences, generating preference vectors that reflect students' knowledge weaknesses, areas of interest, and questioning habits, thereby comprehensively characterizing students' individual characteristics.

[0078] S520 , calculating a first similarity between each candidate answer solution and the professional background feature vector, and a second similarity between the candidate answer solution and the historical question and answer preference feature vector.

[0079] It is understood that for candidate answers, key information such as professional terminology, knowledge points, and theoretical models can be extracted and encoded into semantic vectors. For example, for an answer regarding "The Application of Convolutional Neural Networks in Image Recognition," professional terms such as "convolutional layer," "pooling layer," and "backpropagation" are parsed and converted into high-dimensional semantic vectors using a pre-trained language model (such as BERT). This semantic vector is then compared with the student's professional background feature vector using metrics such as cosine similarity and Euclidean distance. If the student is a senior computer science major whose professional background feature vector demonstrates in-depth knowledge of deep learning, answers that include complex algorithmic derivations and cutting-edge application examples will receive a higher first-level similarity. When calculating the second-level similarity, the candidate answers can be analyzed based on the student's preferences for historical question answering, such as their presentation style, level of detail, and example types. For example, if a student repeatedly asks about the practical application scenarios of a concept in a historical question, answers that include a wealth of practical examples will score higher in the second-level similarity calculation. The system will also use a text classification model to identify the type of answer (such as popular science, academic, and popular). Combined with the adoption of different types of answers in students' historical interactions, it will dynamically adjust the weight of the similarity calculation to more accurately evaluate the degree of match between candidate answer solutions and students' preferences.

[0080] S530: Input the first similarity and the second similarity of each candidate answer solution into a preset deep reasoning screening model, and output a comprehensive matching score of each candidate answer solution.

[0081] It can be understood that this approach can be based on deep learning architectures, such as multi-layer Transformer networks or deep neural networks (DNNs). During the model training phase, a deep reasoning and screening model is trained using a large amount of labeled historical question-and-answer data, which contains information such as student identity, candidate answers, and the final selected target answer. The deep reasoning and screening model uses first and second similarity as input features and uses a multi-head attention mechanism to capture the correlation between the two dimensions. For example, it analyzes whether the influence of historical question-and-answer preferences on answer selection changes when the degree of professional background match is high. By performing nonlinear transformations on the features through a multi-layer perceptron (MLP), the model learns how different feature combinations influence answer quality. During the inference phase, when new candidate answer similarity data is input, the model outputs a comprehensive matching score based on the knowledge learned during training. This comprehensive matching score not only considers the independent effects of professional background and historical question-and-answer preferences, but also incorporates the synergistic effects between the two. For example, for an answer plan that has difficult professional knowledge but an expression style that suits the student's preferences, the deep reasoning screening model will reasonably weigh the professional matching and preference matching based on feedback from similar situations in the training data, and give an objective comprehensive score, thereby achieving a comprehensive evaluation of the candidate answer plans.

[0082] Exemplarily, the training of the deep reasoning screening model can be achieved by constructing a multi-source heterogeneous training data set, including collecting historical question-and-answer interaction records, user identity information and feedback data, integrating professional knowledge bases to construct domain knowledge data, and constructing positive and negative samples in a 1:3 ratio by replacing key information, etc., and then annotating the quality answer evaluation dimensions to extract the user's professional background and historical preference features, the semantic and structural features of the candidate answer solutions, etc.; the model adopts a three-layer architecture of "feature fusion-deep reasoning-scoring output", the input layer splices user features and matching features to form a multi-dimensional tensor, the middle layer captures the correlation between features through multi-layer Transformer or deep neural network combined with multi-head attention mechanism, and learns the synergistic influence pattern of professional adaptability and preference matching, and the output layer generates a comprehensive matching score in the range of 0-1 through the activation function; during the training process, the cross entropy loss function is used as the optimization target, the AdamW optimizer is used in combination with the early stopping strategy to prevent overfitting, and the training stability is improved through batch normalization and gradient clipping. Finally, the hyperparameters are adjusted using the validation set, so that the model can perform deep reasoning and accurate scoring of candidate answer solutions based on the user's personalized features.

[0083] S540: Determine the candidate answer solution with the highest comprehensive matching score as the target answer solution, and show the target answer solution to the student asking the question.

[0084] It can be understood that by comparing the comprehensive matching scores of all candidate answer plans, selecting the target answer plan with the highest score as the final answer, and showing the target answer plan to the students asking the questions, the targeted and authoritative improvement from question understanding, knowledge retrieval to answer generation can be achieved.

[0085] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0086] Corresponding to the intelligent service method based on deep reasoning and knowledge enhancement described in the above embodiment, the embodiment of the present application also provides an intelligent service system based on deep reasoning and knowledge enhancement, and each unit of the system can implement each step of the intelligent service method based on deep reasoning and knowledge enhancement. Figure 2 A structural block diagram of an intelligent service system based on deep reasoning and knowledge enhancement provided by an embodiment of the present application is shown. For the sake of convenience, only the parts related to the embodiment of the present application are shown.

[0087] Reference Figure 2 , the intelligent service system based on deep reasoning and knowledge enhancement includes: An acquisition unit, used to obtain the identity information of the student asking the question and target question information; A parameter unit, configured to construct question-answer feature parameters corresponding to the student asking the question based on the target question information; A first screening unit is configured to screen at least one candidate question and answer unit from a preset question and answer knowledge base based on the question and answer feature parameters; wherein the candidate question and answer unit includes historically similar questions and corresponding answers, and associated document fragments; A generating unit, configured to generate at least one candidate answer solution based on the candidate question and answer unit; wherein any candidate answer solution uses any candidate question and answer unit as a core content carrier; The second screening unit is used to screen the candidate answer solutions based on the identity information using a preset deep reasoning screening model, determine the target answer solution, and present the target answer solution to the student asking the question; wherein, the deep reasoning screening model is a pre-trained machine learning model.

[0088] It should be noted that the information interaction, execution process, etc. between the above-mentioned systems / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0089] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit module can exist physically alone, or two or more unit modules can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0090] The embodiment of the present application also provides an electronic device, Figure 3 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present application. Figure 3 As shown, the electronic device 6 of this embodiment includes: at least one processor 60 ( Figure 3 Only one is shown), at least one memory 61 ( Figure 3 Only one is shown in the figure) and a computer program 62 stored in the at least one memory 61 and executable on the at least one processor 60. When the processor 60 executes the computer program 62, the electronic device 6 implements the steps in any of the above-mentioned embodiments of the intelligent service method based on deep reasoning and knowledge enhancement, or implements the functions of the units in the above-mentioned system embodiments.

[0091] For example, the computer program 62 may be divided into one or more units, which are stored in the memory 61 and executed by the processor 60 to implement the present application. The one or more units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 62 in the electronic device 6.

[0092] The electronic device 6 can be a computing device or terminal device such as a desktop computer, a notebook, a PDA, or a cloud server. The electronic device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will understand that Figure 3 It is only an example of the electronic device 6 and does not constitute a limitation on the electronic device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, buses, etc.

[0093] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0094] In some embodiments, the memory 61 may be an internal storage unit of the electronic device 6, such as a hard drive or memory of the electronic device 6. In other embodiments, the memory 61 may also be an external storage device of the electronic device 6, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped on the electronic device 6. Furthermore, the memory 61 may include both an internal storage unit of the electronic device 6 and an external storage device. The memory 61 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory 61 may also be used to temporarily store data that has been output or is about to be output.

[0095] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0096] An embodiment of the present application provides a computer program product. When the computer program product is run on an electronic device, the electronic device implements the steps of any of the above method embodiments.

[0097] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to an electronic device, a recording medium, computer memory, read-only memory (ROM), random access memory (RAM), an electrical carrier signal, a telecommunications signal, and a software distribution medium. Examples include a USB flash drive, a removable hard drive, a magnetic disk, or an optical disk.

[0098] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0099] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0100] In the embodiments provided in the present application, it should be understood that the disclosed intelligent service system / electronic device and method based on deep reasoning and knowledge enhancement can be implemented in other ways. For example, the above-described embodiments of the intelligent service system / electronic device based on deep reasoning and knowledge enhancement are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0101] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0102] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. An intelligent service method based on deep reasoning and knowledge enhancement, characterized in that: include: Obtain the identity information of the student asking the question and the target question information; Constructing question-answer feature parameters corresponding to the student asking the question according to the target question information; Based on the question and answer feature parameters, at least one candidate question and answer unit is selected from a preset question and answer knowledge base; wherein the candidate question and answer unit includes historically similar questions and corresponding answers, and associated document fragments; Generate at least one candidate answer solution based on the candidate question and answer unit; wherein any candidate answer solution uses any candidate question and answer unit as a core content carrier; The candidate answer solutions are screened using a preset deep reasoning screening model based on the identity information, a target answer solution is determined, and the target answer solution is presented to the student asking the question; wherein the deep reasoning screening model is a pre-trained machine learning model.

2. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 1, characterized in that: Before obtaining the identity information of the student asking the question and the target question information, the method further includes: Obtain documents related to question and answer content, and build a question and answer knowledge base based on the documents related to the question and answer content; wherein the question and answer knowledge base includes a first knowledge warehouse built based on domain labels, and a second knowledge warehouse built based on semantic feature vectors.

3. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 2, characterized in that: The step of obtaining documents related to the question and answer content and constructing a question and answer knowledge base based on the documents related to the question and answer content includes: Obtaining documents related to the question and answer content, and extracting seed question and answer units marked with domain labels from the documents related to the question and answer content; Performing semantic analysis on each of the seed question-answering units to generate a feature vector corresponding to the seed question-answering unit; Building a semantic index database based on the feature vector, and automatically assigning domain labels to question-answer units that are not labeled with domain labels using the semantic index database; All seed question-answering units are clustered based on the domain labels to form a first knowledge warehouse.

4. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 3, characterized in that: Automatically assigning domain labels to question-answer units that are not labeled with domain labels using the semantic index database includes: Perform semantic analysis on question-answer units without domain labels to generate target feature vectors; Calculating the similarity between the target feature vector and each feature vector in the semantic index database, and screening out candidate seed question-answering units whose similarity exceeds a threshold; Performing semantic consistency check on the candidate seed question and answer units to determine the target seed question and answer unit with the highest matching degree; Assigning the domain label of the target seed question-answering unit to the question-answering unit.

5. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 1, characterized in that: The step of constructing question-answer feature parameters corresponding to the student asking the question according to the target question information includes: Performing semantic analysis based on the target question information to determine a semantic knowledge subgraph corresponding to the target question information and a semantic index segment of the semantic knowledge subgraph; Constructing a semantic adjacency tensor corresponding to the semantic knowledge subgraph based on the semantic index fragment; Topologically sorting the semantic adjacency tensor to generate a semantic subgraph dependency tree; wherein the semantic subgraph dependency tree explicitly expresses the hierarchical dependency relationship between semantic knowledge subgraphs in a tree structure; The semantic subgraph dependency tree is hierarchically decomposed to determine question-answer feature parameters corresponding to the student asking the question.

6. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 2, characterized in that: The question-answering feature parameters include domain labels and semantic feature vectors; The step of selecting at least one candidate question and answer unit from a preset question and answer knowledge base based on the question and answer feature parameters includes: Matching a corresponding first question-answer unit set from the first knowledge warehouse according to the domain label, and retrieving a corresponding second question-answer unit set from the second knowledge warehouse according to the semantic feature vector; respectively determining a first matching coefficient list for the first question-and-answer unit set and a second matching coefficient list for the second question-and-answer unit set; Cross-merging the first question-and-answer unit set and the second question-and-answer unit set based on the first matching coefficient list and the second matching coefficient list to obtain a cross-question-and-answer unit set; wherein the cross-question-and-answer unit set includes at least one cross-question-and-answer unit; Each cross-question and answer unit in the cross-question and answer unit set is determined as a candidate question and answer unit.

7. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 6, characterized in that: The determining of a first matching coefficient list for the first question-and-answer unit set and a second matching coefficient list for the second question-and-answer unit set respectively includes: Calculating the degree of matching between each first question and answer unit in the first question and answer unit set and the target question information to obtain a first matching coefficient for each first question and answer unit, and forming a first matching coefficient list from all first matching coefficients; wherein the first matching coefficient is calculated based on the domain label, the historical adoption rate score, and the historical follow-up rate; Calculate the degree of matching between each second question and answer unit in the second question and answer unit set and the target question information, obtain the first matching coefficient of each second question and answer unit, and form all the second matching coefficients into a second matching coefficient list; wherein, the second matching coefficient is obtained by calculating the cosine similarity and the question and answer completeness.

8. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 5, characterized in that: Generating at least one candidate answer solution based on the candidate question-answer unit includes: For each candidate question-answering unit, calculating the correlation between the candidate question-answering unit and the hierarchical nodes of the semantic subgraph dependency tree, and extracting node document fragments corresponding to the hierarchical nodes whose correlation is greater than a threshold; Based on the extracted node document fragments, the candidate question and answer units are expanded and optimized through natural language generation to generate candidate answer solutions.

9. The intelligent service method based on deep reasoning and knowledge enhancement according to claim 1, characterized in that: The candidate answer solutions are screened using a preset deep reasoning screening model based on the identity information to determine a target answer solution, and the target answer solution is presented to the student asking the question; wherein the deep reasoning screening model is a pre-trained machine learning model, including: Extracting the student's professional background feature vector and historical question-answering preference feature vector based on the identity information; Calculating a first similarity between each candidate answer solution and the professional background feature vector, and a second similarity between each candidate answer solution and the historical question and answer preference feature vector; Inputting the first similarity and the second similarity of each candidate answer solution into a preset deep reasoning screening model, and outputting a comprehensive matching score for each candidate answer solution; The candidate answer solution with the highest comprehensive matching score is determined as the target answer solution, and the target answer solution is presented to the student asking the question.

10. An intelligent service system based on deep reasoning and knowledge enhancement, characterized in that: include: An acquisition unit, used to obtain the identity information of the student asking the question and target question information; A parameter unit, configured to construct question-answer feature parameters corresponding to the student asking the question based on the target question information; A first screening unit is configured to screen at least one candidate question and answer unit from a preset question and answer knowledge base based on the question and answer feature parameters; wherein the candidate question and answer unit includes historically similar questions and corresponding answers, and associated document fragments; A generating unit, configured to generate at least one candidate answer solution based on the candidate question and answer unit; wherein any candidate answer solution uses any candidate question and answer unit as a core content carrier; The second screening unit is used to screen the candidate answer solutions based on the identity information using a preset deep reasoning screening model, determine the target answer solution, and present the target answer solution to the student asking the question; wherein, the deep reasoning screening model is a pre-trained machine learning model.

Images