Entity disambiguation processing method and device in insurance field, equipment, storage medium and program product

Abstract

The invention relates to the technical field of natural language processing, provides an entity disambiguation processing method and device in the field of insurance, equipment, a storage medium and a program product, and can improve the entity disambiguation accuracy of an insurance text. The method comprises the steps of obtaining to-be-disambiguated entities in an insurance text; determining a candidate entity set matched with the entity to be disambiguated in the insurance field knowledge base; determining a context text of the to-be-disambiguated entity in the insurance text, and performing insurance service feature extraction according to the to-be-disambiguated entity and the context text to obtain insurance service features; for each candidate entity in the candidate entity set, fusing the embedded vector of the candidate entity in the insurance field knowledge base, the insurance business feature and the embedded vector of the context text to obtain a fusion vector corresponding to the candidate entity; and according to the matching degree between the fusion vector and the to-be-disambiguated entity, sorting the candidate entities in the candidate entity set, and determining a target candidate entity to obtain a disambiguation result.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, and in particular to a method, apparatus, computer equipment, storage medium, and computer program product for entity disambiguation processing in the field of insurance. Background Technology

[0002] In insurance intelligent customer service Q&A scenarios, a large number of business scenarios involve unstructured text data, such as insurance users' claim descriptions, diagnoses, medical records, and claim materials. These text materials contain a large number of professional terms, abbreviations, synonyms / near-synonyms, and vague or incomplete entities, resulting in entity ambiguity issues such as homonyms, synonyms, ambiguous references, asymmetry between full abbreviations, and regional differences.

[0003] Entity disambiguation methods in related technologies can be based on general knowledge bases and open-domain knowledge graphs. This method performs well in general domains, but it has obvious shortcomings in highly specialized scenarios such as intelligent insurance customer service, where the accuracy of entity disambiguation decreases. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, storage medium, and computer program product for entity disambiguation processing in the field of insurance, addressing the aforementioned technical problems.

[0005] This application provides a method for entity disambiguation processing in the field of insurance, the method comprising:

[0006] Retrieve the entity to be disambiguated from the insurance document;

[0007] In the insurance domain knowledge base, determine a set of candidate entities that match the entity to be disambiguated;

[0008] In the insurance text, the context text of the entity to be disambiguated is determined, and insurance business features are extracted based on the entity to be disambiguated and the context text to obtain insurance business features;

[0009] For each candidate entity in the candidate entity set, the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features and the context text are fused to obtain the fused vector corresponding to the candidate entity;

[0010] Based on the matching degree between the fusion vector and the entity to be disambiguated, the candidate entities in the candidate entity set are sorted to determine the target candidate entity, so as to obtain the disambiguation result.

[0011] This application provides a physical disambiguation processing device in the field of insurance, the device comprising:

[0012] The entity identification module is used to obtain the entities to be disambiguated in the insurance text;

[0013] The candidate entity recall module is used to determine a set of candidate entities that match the entity to be disambiguated in the insurance domain knowledge base;

[0014] The business feature extraction module is used to determine the context text of the entity to be disambiguated in the insurance text, and to extract insurance business features based on the entity to be disambiguated and the context text to obtain insurance business features.

[0015] The fusion processing module is used to fuse the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features and the embedding vector of the context text for each candidate entity in the candidate entity set, so as to obtain the fusion vector corresponding to the candidate entity.

[0016] The disambiguation result determination module is used to sort the candidate entities in the candidate entity set according to the matching degree between the fusion vector and the entity to be disambiguated, and determine the target candidate entity to obtain the disambiguation result.

[0017] This application provides a computer device, including a memory and a processor, wherein the memory stores a computer program and the processor executes the above-described method.

[0018] This application provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor using the methods described above.

[0019] This application provides a computer program product having a computer program stored thereon, the computer program being executed by a processor using the above-described method.

[0020] The aforementioned entity disambiguation processing method, apparatus, computer equipment, storage medium, and computer program product in the insurance field acquire the entity to be disambiguated from the insurance text; determine a set of candidate entities matching the entity to be disambiguated in an insurance domain knowledge base; determine the context text of the entity to be disambiguated in the insurance text; extract insurance business features based on the entity to be disambiguated and the context text to obtain insurance business features; for each candidate entity in the candidate entity set, fuse the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features, and the embedding vector of the context text to obtain a fused vector corresponding to the candidate entity; and sort the candidate entities in the candidate entity set according to the matching degree between the fused vector and the entity to be disambiguated to determine the target candidate entity, thereby obtaining the disambiguation result. The solution provided in this application leverages an insurance domain knowledge base for entity disambiguation. This knowledge base can include professional entities related to insurance, such as specific disease codes, insurance product names, underwriting rules, and exclusion clauses, enhancing the connection with insurance semantics and enabling a more accurate understanding of insurance meaning. Furthermore, insurance texts may contain abbreviations, vague references, homonyms, and synonyms that require disambiguation. With the support of the insurance domain knowledge base, semantic equivalence, hierarchical relationships, and exclusion relationships between entities can be established, thereby improving the accuracy of entity disambiguation. Additionally, the embedding vectors of candidate entities in the insurance domain knowledge base, insurance business features, and the embedding vectors of the context text of the entity to be disambiguated are fused. The matching degree between this fused result and the entity to be disambiguated is then used for ranking. Introducing insurance business features and contextual semantics during the ranking stage ensures that the target candidate entities determined based on the ranking results conform to business rules, further improving the accuracy of entity disambiguation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating an entity disambiguation processing method in the insurance field, as shown in one embodiment.

[0023] Figure 2 This is a flowchart illustrating the process of determining the entity to be disambiguated in one embodiment;

[0024] Figure 3 This is another flowchart illustrating an entity disambiguation processing method in the insurance field, as shown in one embodiment.

[0025] Figure 4This is a structural block diagram of a physical disambiguation processing device in the field of insurance, as shown in one embodiment.

[0026] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various objects, but these objects are not limited by these terms. These terms are only used to distinguish the first object from the second object. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0029] The entity disambiguation processing method in the insurance field provided in this application can be executed by computer equipment, and the method may include... Figure 1 The steps shown are as follows:

[0030] Step S101: Obtain the entity to be disambiguated in the insurance text.

[0031] Text in the insurance field can be referred to as insurance text. Insurance text can include, but is not limited to, texts of insurance user inquiries, customer service hotline call logs, insurance policies, and claim forms. After obtaining the insurance text, data preprocessing can be performed, including but not limited to cleaning, sentence segmentation, word segmentation, stop word removal, and part-of-speech tagging.

[0032] Step S102: Determine the set of candidate entities that match the entity to be disambiguated in the insurance domain knowledge base.

[0033] An insurance domain knowledge base is a specialized knowledge base for the insurance field. It integrates structured data from historical and general data within the insurance sector, defines core concepts, attributes, and relationships within the insurance domain, and constructs a domain ontology and knowledge graph, thus generating the insurance domain knowledge base. The structured data from historical and general data can include, but is not limited to, insurance product catalogs, terms and conditions libraries, disease coding libraries, drug catalogs, hospital and doctor directories, underwriting and claims rule libraries, and customer profiles. The insurance domain knowledge base can be used to store structured insurance entities (which can be simply referred to as entities). Each entity can contain data such as entity type, entity ID, standard name, entity attributes, entity relationships, and embedding vectors.

[0034] entity M to be disambiguated i Taking an example, candidate entity recall can be performed based on an insurance domain knowledge base. Multiple search strategies are employed within this pre-built knowledge base to identify entities M that are likely to be disambiguated. i Several matched candidate entities can form a corresponding candidate entity set, which can be represented as Candidates(M). i )={C1,C2,…,C k}

[0035] Step S103: Determine the context text of the entity to be disambiguated in the insurance text, and extract insurance business features based on the entity to be disambiguated and the context text to obtain insurance business features.

[0036] entity M to be disambiguated i Taking the entity M to be disambiguated as an example, we can introduce it as an example. i The preceding and following 50 characters serve as a context window. Within the insurance text, the entity M to be disambiguated is located. i The content before and after the 50 words, the office as the entity M to be eliminated i The context text.

[0037] After obtaining the context text, the context text and the entity M to be disambiguated can be compared. i The result of extracting contextual semantic features can be called the embedding vector V of the contextual text. context It can also be based on the entity M to be disambiguated. i Insurance business features V are extracted from the context text. business Insurance business characteristics can include multiple dimensions of business characteristics.

[0038] Step S104: For each candidate entity in the candidate entity set, fuse the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of insurance business features and the context text to obtain the fused vector corresponding to the candidate entity.

[0039] Candidates(M) is a set of candidate entities. i )={C1,C2,…,C k Candidate entity C in} j For example, candidate entity C can be identified in the insurance domain knowledge base. j Embedding vector V Cj_kg This embedding vector can also be called a relation representation, knowledge representation, knowledge graph embedding vector, etc. It can be used for candidate entity C. j Embedding vector V Cj_kg Insurance Business Characteristics V business Embedding vector V with context text context The fusion process yields candidate entity C. j The corresponding fusion vector V fused .

[0040] Step S105: Based on the matching degree between the fusion vector and the entity to be disambiguated, sort the candidate entities in the candidate entity set to determine the target candidate entity and obtain the disambiguation result.

[0041] After obtaining candidate entity C j The corresponding fusion vector V fused Then, the fusion vector V can be processed. fused Disambiguated entity M i The matching degree between them is calculated to obtain the matching degree with the candidate entity C. j The corresponding matching degree can be denoted as Score(M) i C j Score(M) i C j )∈[0,1].

[0042] Following the aforementioned method, the candidate entity set Candidates(M) can be calculated. i )={C1,C2,…,C k The matching degree corresponding to each candidate entity in} can be used to classify the candidate entity set Candidates(M) according to the relative magnitude of the matching degree. i )={C1,C2,…,C k The candidate entities in the list are sorted in descending order, so that the top few candidate entities (such as the first candidate entity) can be identified. These candidate entities can be used as target candidate entities, and the target candidate entities can be used as the disambiguation result.

[0043] In the aforementioned entity disambiguation processing method in the insurance field, the following steps are taken: First, the entity to be disambiguated in the insurance text is obtained. Second, a set of candidate entities matching the entity to be disambiguated is determined in the insurance domain knowledge base. Third, the context text of the entity to be disambiguated is determined in the insurance text, and insurance business features are extracted based on the entity to be disambiguated and the context text to obtain insurance business features. Fourth, for each candidate entity in the candidate entity set, the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features, and the embedding vector of the context text are fused to obtain a fused vector corresponding to the candidate entity. Fifth, based on the matching degree between the fused vector and the entity to be disambiguated, the candidate entities in the candidate entity set are sorted to determine the target candidate entity, thus obtaining the disambiguation result. The solution provided in this application leverages an insurance domain knowledge base for entity disambiguation. This knowledge base can include professional entities related to insurance, such as specific disease codes, insurance product names, underwriting rules, and exclusion clauses, enhancing the connection with insurance semantics and enabling a more accurate understanding of insurance meaning. Furthermore, insurance texts may contain abbreviations, vague references, homonyms, and synonyms that require disambiguation. With the support of the insurance domain knowledge base, semantic equivalence, hierarchical relationships, and exclusion relationships between entities can be established, thereby improving the accuracy of entity disambiguation. Additionally, the embedding vectors of candidate entities in the insurance domain knowledge base, insurance business features, and the embedding vectors of the context text of the entity to be disambiguated are fused. The matching degree between this fused result and the entity to be disambiguated is then used for ranking. Introducing insurance business features and contextual semantics during the ranking stage ensures that the target candidate entities determined based on the ranking results conform to business rules, further improving the accuracy of entity disambiguation.

[0044] In one embodiment, such as Figure 2 As shown, the entity to be disambiguated in the insurance text is obtained, including:

[0045] Step S201: Input the insurance text into the pre-built insurance entity recognition model; the insurance entity recognition model is obtained by adjusting the natural language processing model based on the insurance dataset; Step S202: Obtain the entity to be disambiguated based on the output of the insurance entity recognition model.

[0046] Insurance datasets can include, but are not limited to, anonymized insurance policies, claims reports, and customer service dialogues. Based on these datasets, natural language processing (NLP) models are adaptively fine-tuned using insurance domain knowledge. This enables the pre-trained NLP model to possess insurance semantic understanding capabilities, resulting in an insurance entity recognition model. The insurance entity recognition model can utilize an attention mechanism to focus on the context most relevant to disambiguating the target entity and related entities and attributes in the knowledge base. The NLP model can include, but is not limited to, the BERT model. BERT stands for Bidirectional Encoder Representations from Transformers.

[0047] Insurance text can be input into a pre-built insurance entity recognition model. The insurance entity recognition model processes the input insurance text and can identify ambiguous entities to be disambiguated within the text.

[0048] This embodiment uses an insurance dataset to adjust the natural language processing model, enabling the pre-trained natural language processing model to have the ability to understand insurance semantics and accurately identify entities to be disambiguated in insurance texts.

[0049] In one embodiment, determining a set of candidate entities that match the entity to be disambiguated in an insurance domain knowledge base includes:

[0050] When the standard name corresponding to an entity in the insurance domain knowledge base matches the entity to be disambiguated, that entity is identified as a candidate entity matching the entity to be disambiguated; when the distance required to edit the entity to be disambiguated into an entity in the insurance domain knowledge base is less than a set distance threshold, that entity is identified as a candidate entity matching the entity to be disambiguated; when the encoding information of the entity to be disambiguated matches the encoding information corresponding to an entity in the insurance domain knowledge base, that entity is identified as a candidate entity matching the entity to be disambiguated; based on the candidate entities matching the entity to be disambiguated, a set of candidate entities is obtained.

[0051] entity M to be disambiguated i Taking this as an example, candidate entity recall can be performed based on an insurance domain knowledge base, employing a multi-strategy retrieval method within the pre-built insurance domain knowledge base.

[0052] The first retrieval strategy can be an exact match strategy, which targets entities that can match the standard name or alias; specifically, it can target the entity M to be disambiguated. i Matching the standard name corresponding to each entity in the insurance domain knowledge base; if an entity matches the entity to be disambiguated M... iIf matched, then the entity can be used as the entity to be disambiguated, M. i Matching candidate entities; specifically, the entity to be disambiguated M can also be... i Matching with the alias corresponding to each entity in the insurance domain knowledge base; if an entity matches the entity to be disambiguated M... i If matched, then the entity can be used as the entity to be disambiguated, M. i Matching candidate entities.

[0053] The second retrieval strategy can be a fuzzy matching strategy, which aims to ensure that the edit distance does not exceed a set distance threshold; specifically, for each entity in the insurance domain knowledge base, the entity M to be disambiguated can be determined. i The number of editing steps required to compile the entity into the insurance domain knowledge base can be referred to as the distance. If this distance is less than a set distance threshold, the corresponding entity is identified as a candidate entity matching the entity to be disambiguated. For example, entity M to be disambiguated... i If an entity can be obtained through two or fewer editing steps, then it can be determined that the entity belongs to the entity M to be disambiguated. i The approximate string of the entity can be identified as a candidate entity that matches the entity to be disambiguated.

[0054] The third retrieval strategy can be an encoding matching strategy. If an entity contains encoding information, it can be retrieved based on that encoding information. Specifically, if the entity to be disambiguated contains encoding information, that encoding information can be matched with the encoding information corresponding to each entity in the insurance domain knowledge base; if the encoding information of a certain entity matches the encoding information of the entity to be disambiguated, that entity can be identified as a candidate entity that matches the entity to be disambiguated.

[0055] Based on several candidate entities that match the entity to be disambiguated, a set of candidate entities can be obtained.

[0056] This embodiment can employ multi-strategy retrieval to recall candidate entities in a pre-built insurance domain knowledge base, which can avoid omissions and improve disambiguation accuracy.

[0057] In one embodiment, insurance business features are extracted based on the entity to be disambiguated and the context text to obtain insurance business features, including:

[0058] Insurance segment type identification is performed on the undisambiguated entity and the context text to obtain insurance segment category labels; insurance liability association identification is performed on the undisambiguated entity and the context text to obtain insurance liability labels; insurance user attribute association identification is performed on the undisambiguated entity and the context text to obtain insurance user attribute feature vectors; and insurance business features are obtained based on the insurance segment category labels, insurance liability labels, and insurance user attribute feature vectors.

[0059] After obtaining the entity to be disambiguated and the context text, multi-directional business features can be obtained based on the entity to be disambiguated and the context text, thus forming the insurance business feature V. business .

[0060] Insurance paragraph type identification can be performed. During this process, based on keywords and a classification model, it can determine whether the entity to be disambiguated and the surrounding text involve areas such as "past medical history," "current diagnosis," or "disclaimer," thereby obtaining the paragraph category label T. segment .

[0061] Insurance liability association identification can be performed. During this process, rules and NER models can be used to identify whether the entity to be disambiguated and the context text contain keywords such as "critical illness insurance" and "accidental injury liability," thereby obtaining the insurance liability label T. coverage NER stands for Named Entity Recognition.

[0062] It can perform insurance user attribute association identification. In this process, it can obtain the profile of the insurance user from the system, including their age, occupation, and insurance products, thereby obtaining the insurance user attribute feature vector V. customer .

[0063] Based on the paragraph category label T segment Insurance liability label T coverage Insurance user attribute feature vector V customer Combining to form insurance business characteristics V business , can be represented as V business =[T segment T coverage V customer ].

[0064] This embodiment, based on insurance segment type recognition, insurance liability association recognition, and insurance user attribute association recognition, can extract business features in three dimensions, thereby obtaining insurance business features. This allows the target candidate entities determined based on the ranking results to conform to business rules, thus improving the accuracy of entity disambiguation.

[0065] In one embodiment, the embedding vectors of candidate entities in the insurance domain knowledge base, insurance business features, and contextual text are fused to obtain the fused vector corresponding to the candidate entity, including:

[0066] The embedding vectors of candidate entities in the insurance domain knowledge base, the embedding vectors of insurance business features and context text are weighted and summed; based on the weighted summation result, the fusion vector corresponding to the candidate entity is obtained.

[0067] Candidates(M) is a set of candidate entities. i )={C1,C2,…,C k Candidate entity C in} j For example, a dynamic weighted fusion method using an attention mechanism can be used to perform fusion on candidate entity C. j Embedding vector V Cj_kg Insurance Business Characteristics V business Embedding vector V with context text context Perform a weighted summation, and the candidate entity C can be obtained from the weighted summation result. j The corresponding fusion vector V fused , can be represented as V fused =θ(V context V business V Cj_kg ), where θ(•) can be expressed as:

[0068] Attention_Weight=Softmax(W1·V context +W2·V business +W3·V Cj_kg );

[0069] Accordingly, V fused It can be represented as:

[0070] V fused =Attention_Weight[0]·V context +Attention_Weight[1]·V business +Attention_Weight[2]·V Cj_kg .

[0071] Softmax can represent a normalized exponential function, and Attention_Weight can represent the weights determined through the attention mechanism.

[0072] This embodiment uses a weighted summation method to fuse the embedding vectors of candidate entities in the insurance domain knowledge base, insurance business features, and contextual text, resulting in a more accurate fused vector.

[0073] In one embodiment, the candidate entities in the candidate entity set are sorted according to the matching degree between the fusion vector and the entity to be disambiguated, including:

[0074] The fusion vector and the entity to be disambiguated are input into a pre-built matching degree evaluation model; the matching degree between the fusion vector and the entity to be disambiguated is obtained based on the output of the matching degree evaluation model; and the candidate entities in the candidate entity set are sorted according to the relative size of the matching degree.

[0075] After obtaining candidate entity C j The corresponding fusion vector V fused Then, the fusion vector V can be... fused Disambiguated entity M i The input is fed into a pre-built matching evaluation model. The fusion vector V is obtained from the output of the matching evaluation model. fused M between the entity to be disambiguated i The degree of matching, which can be called the matching score.

[0076] Following the aforementioned method, the candidate entity set Candidates(M) can be calculated. i )={C1,C2,…,C k The matching degree corresponding to each candidate entity in} can be used to classify the candidate entity set Candidates(M) according to the relative magnitude of the matching degree. i )={C1,C2,…,C k The candidate entities in the list are sorted in descending order to determine the candidate entity with the highest matching degree. This candidate entity can be used as the target candidate entity and the target candidate entity can be used as the disambiguation result.

[0077] In this embodiment, the matching degree evaluation model can obtain a more accurate matching degree between the fusion vector and the entity to be disambiguated, thereby improving the disambiguation accuracy.

[0078] To better understand the above method, the following details an application example of the entity disambiguation processing method in the insurance field of this application. The solution provided in this application example deeply integrates insurance domain knowledge, contextual semantics, and business rules, which can improve the accuracy and recall rate of entity recognition in insurance text. This application example constructs an insurance domain knowledge base, which provides strong constraints and rich clues for the overall disambiguation process. Then, using the insurance domain knowledge base, an entity disambiguation solution integrating insurance domain knowledge is constructed, which can include the following three parts:

[0079] 1. Entity Recognition Process: Using anonymized insurance policies, claims reports, and customer service dialogues, the BERT model is adaptively fine-tuned to incorporate insurance domain knowledge, enabling it to possess insurance semantic understanding capabilities, resulting in an insurance entity recognition model. This model performs entity recognition by employing an attention mechanism, focusing on the context most relevant to disambiguation of the target entity and related entities and attributes in the knowledge base. The insurance entity recognition model can process input insurance text and identify ambiguous entities to be disambiguated.

[0080] 2. Candidate Entity Recall Process: Based on the entities to be disambiguated extracted during the entity recognition process, a search is performed in the constructed insurance domain knowledge base to recall a set of candidate entities related to the entities to be disambiguated. The set of candidate entities may contain one or more candidate entities.

[0081] 3. Candidate Entity Ranking and Disambiguation Process: Contextual features related to the entities to be disambiguated are extracted. These features are then combined with insurance business characteristics and the embedding vectors of each candidate entity in the insurance domain knowledge base to perform multi-source data fusion, yielding a fusion result. The matching degree between the fusion result and the entities to be disambiguated is calculated. Based on the matching degree, the candidate entities in the candidate entity set are sorted in descending order. The top k candidate entities in the sorted result are identified as the target candidate entities, thus obtaining the disambiguation result.

[0082] The solution provided in this application example involves data preprocessing and building a knowledge base for the insurance field.

[0083] Data preprocessing: Insurance texts may include user inquiries, customer service hotline call logs, policy claims, etc. Data preprocessing can be performed on these texts, including data cleaning steps such as word segmentation, stop word removal, and part-of-speech tagging.

[0084] Building a knowledge base for the insurance field: This involves integrating structured data from historical and general data in the insurance field, defining core concepts, attributes, and relationships within the insurance domain, and constructing a domain ontology and knowledge graph.

[0085] In the solution provided in this application example, entity disambiguation can be performed based on entity recognition, candidate entity recall, and candidate entity ranking. This solution may include the following units:

[0086] Text input and preprocessing unit: Used to receive insurance text (such as original insurance user dialogue text) and perform data preprocessing such as cleaning, sentence segmentation, word segmentation, and part-of-speech tagging on the insurance text.

[0087] Entity recognition unit: Using rule templates and insurance entity recognition models, it identifies entities to be disambiguated in insurance texts.

[0088] Insurance domain knowledge base unit: Stores structured insurance entity data, including entity type, entity attributes, entity relationships, and embedding vectors.

[0089] Candidate Entity Recall Unit: Based on the mentioned surface string, retrieve a set of matching candidate entities from the insurance domain knowledge base.

[0090] The disambiguation engine unit integrates contextual features, insurance business features, and the embedding vectors (which can be called knowledge representations) of candidate entities. It uses a ranking model to calculate the matching degree corresponding to each candidate entity, sorts them in descending order according to the matching degree, and selects the top-ranked candidate entities as the disambiguation results.

[0091] Output Unit: Outputs structured disambiguation results, which may include the entities to be disambiguated, target candidate entities, and confidence levels in the insurance text.

[0092] This application example may include Figure 3 The steps shown are as follows:

[0093] Step S301, Receive input data: Input data may include the user-inputted claim text, electronic policy, claim application form and other intelligent customer service dialogue logs. The insurance text can be obtained based on the input data.

[0094] Step S302, text data preprocessing: Clean the insurance text to remove irrelevant symbols, HTML tags in the card information, emoticons and garbled characters, and then segment the insurance text into words.

[0095] Step S303, Entity Recognition Process: This step combines rule-based templates with an insurance entity recognition model. For the rule templates, a structured data construction approach can be used to create them. For the insurance entity recognition model, an insurance dataset can be used to fine-tune the natural language processing (NLP) model, resulting in a more suitable NLP model for entity recognition tasks in the insurance domain. This model can be called the insurance entity recognition model. The insurance entity recognition model can identify entities such as diseases, drugs, and insurance names in the insurance domain, thereby improving the model's adaptability and enhancing its entity recognition accuracy. The fine-tuned NER model can be used to recognize entities such as diseases, drugs, institutions, and products.

[0096] Step S304, Candidate Entity Recall Process: For the entity M to be disambiguated i It can extract the entity M to be disambiguated. iBased on the characteristics of the insurance domain knowledge base, a multi-strategy retrieval approach is adopted: the first retrieval strategy can be a precise matching strategy, which can completely match the standard name or alias of the target; the second retrieval strategy can be a fuzzy matching strategy, which selects the appropriate match for the entity M to be disambiguated. i In other words, the edit distance is no more than 2 for approximate strings; the third retrieval strategy can be an encoding matching strategy, if the entity to be disambiguated M i If the code contains encoding information, then retrieval can be performed based on that encoding information. After the retrieval is complete, a set of candidate entities is obtained, which can be denoted as Candidates(M). i )={C1,C2,…,C k Each candidate entity can contain an entity ID, standard name, entity attributes, entity relationships, and embedding vector.

[0097] Step S305, Context and Business Feature Extraction Process: This can be performed on the entity M to be disambiguated. i Contextual semantic features are extracted from the insurance text and its surrounding text. Specifically, the entity M to be disambiguated in the insurance text can be extracted. i The preceding and following 50 characters serve as contextual text, defining the entity M to be disambiguated. i The contextual text is input into a BERT model fine-tuned with insurance domain knowledge, which then determines the disambiguation of the entity M. i By performing vector computation with the context text, a corresponding 768-dimensional embedding vector V can be output. context The embedding vector belongs to the contextual semantic features.

[0098] It can be based on the entity M to be disambiguated i Insurance segment type identification is performed in conjunction with the context text. During this identification process, keywords and classification models can be used to determine whether the entity to be disambiguated and the context text involve areas such as "past medical history," "current diagnosis," or "disclaimer," thereby obtaining the insurance segment category label T. segment .

[0099] It can be based on the entity M to be disambiguated i To identify insurance liability associations with the context text, rules and NER models can be used to identify the entity to be disambiguated and whether the context text contains keywords such as "critical illness insurance" and "accidental injury liability," thereby obtaining the insurance liability label T. coverage .

[0100] It is possible to perform disambiguation based on entity M. iIn the contextual text-based insurance user attribute association identification, the system can retrieve the insurance user's profile, including their age, occupation, and insured products, thereby obtaining the insurance user attribute feature vector V. customer .

[0101] Based on the paragraph category label T segment Insurance liability label T coverage Insurance user attribute feature vector V customer Combining to form insurance business characteristics V business , can be represented as V business =[T segment T coverage V customer ].

[0102] Step S306, Candidate Entity Ranking and Disambiguation Decision Process: The candidate entity set Candidates(M) can be ranked and disambiguated. i )={C1,C2,…,C k Each candidate entity in the algorithm undergoes steps such as embedding vector acquisition, feature fusion, matching degree calculation, and ranking decision. In the embedding vector acquisition step, candidate entity C can be obtained from the insurance domain knowledge base. j Embedding vector V Cj_kg In the feature fusion step, the embedding vector V can be adjusted according to the attention mechanism. Cj_kg Insurance Business Characteristics V business Embedding vector V with context text context Performing a weighted summation, the resulting weighted summation can be used to select candidate entity C. j The corresponding fusion vector V fused In the matching degree calculation step, candidate entity C can be calculated. j The corresponding fusion vector V fused Disambiguated entity M i The degree of matching between them is used to obtain the match with the candidate entity C. j The corresponding matching degree can be denoted as Score(M) i C j Score(M) i C j )∈[0,1].

[0103] Based on the relative magnitude of the matching degree, the candidate entity set Candidates(M) can be classified. i )={C1,C2,…,C k The candidate entities in the list are sorted in descending order, so that the top few candidate entities (such as the first candidate entity) can be identified. These candidate entities can be used as target candidate entities, and the target candidate entities can be used as the disambiguation result.

[0104] The solution provided in this application example has strong domain adaptability. The insurance domain knowledge base can include specialized entities in the insurance field, such as specific disease codes, insurance product names, underwriting rules, and exclusion clauses, avoiding a disconnect from insurance language. By combining the insurance domain knowledge base with a domain fine-tuning model, insurance semantics can be accurately understood. Supported by the insurance domain knowledge graph in the knowledge base, semantic equivalence, hierarchical, and exclusion relationships between entities can be established. Furthermore, it can be tightly integrated with business logic, incorporating insurance liability, exclusions, and customer attributes during the sorting stage. Business characteristics enable ranking results to better match business logic, making decisions more consistent with business rules and improving the accuracy of insurance entity disambiguation. This supports the actual needs of high-precision application scenarios such as intelligent underwriting and claims consultation during question-and-answer processes. Furthermore, a multi-stage retrieval strategy can be employed in the recall phase to minimize the omission of candidate entities and preserve their integrity. In the ranking phase, the F1 score can be used to determine the matching degree evaluation model, thereby improving the accuracy of the matching degree. The F1 score, also known as the F1-Score, is the harmonic mean of precision and recall. The F1 score can be used to comprehensively evaluate model performance, simultaneously considering both accuracy and completeness. The solution in this application example adopts a modular design, supports dynamic updates to the knowledge base and continuous model optimization, and is scalable and maintainable.

[0105] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0106] Based on the same inventive concept, this application also provides an entity disambiguation processing apparatus for implementing the entity disambiguation processing method in the insurance field described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations of one or more embodiments of the entity disambiguation processing apparatus in the insurance field provided below can be found in the limitations of the entity disambiguation processing method in the insurance field described above, and will not be repeated here.

[0107] In one embodiment, such as Figure 4 As shown, a physical disambiguation processing device in the field of insurance is provided, comprising:

[0108] Entity determination module 401 is used to obtain the entities to be disambiguated in the insurance text;

[0109] The candidate entity recall module 402 is used to determine a set of candidate entities that match the entity to be disambiguated in the insurance domain knowledge base;

[0110] The business feature extraction module 403 is used to determine the context text of the entity to be disambiguated in the insurance text, and to extract insurance business features based on the entity to be disambiguated and the context text to obtain insurance business features.

[0111] The fusion processing module 404 is used to fuse the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features and the embedding vector of the context text for each candidate entity in the candidate entity set, so as to obtain the fusion vector corresponding to the candidate entity.

[0112] The disambiguation result determination module 405 is used to sort the candidate entities in the candidate entity set according to the matching degree between the fusion vector and the entity to be disambiguated, and determine the target candidate entity to obtain the disambiguation result.

[0113] In one embodiment, the entity determination module 401 is used for:

[0114] The insurance text is input into a pre-built insurance entity recognition model; the insurance entity recognition model is obtained by adjusting a natural language processing model based on an insurance dataset; the entity to be disambiguated is obtained based on the output of the insurance entity recognition model.

[0115] In one embodiment, the candidate entity recall module 402 is used for:

[0116] When the standard name corresponding to an entity in the insurance domain knowledge base matches the entity to be disambiguated, the entity is identified as a candidate entity matching the entity to be disambiguated; when the distance required for the entity to be disambiguated to be edited into an entity in the insurance domain knowledge base is less than a set distance threshold, the entity is identified as a candidate entity matching the entity to be disambiguated; when the encoding information of the entity to be disambiguated matches the encoding information corresponding to an entity in the insurance domain knowledge base, the entity is identified as a candidate entity matching the entity to be disambiguated; based on the candidate entities matching the entity to be disambiguated, the candidate entity set is obtained.

[0117] In one embodiment, the business feature extraction module 403 is used for:

[0118] The insurance segment type is identified by performing insurance segment type identification on the entity to be disambiguated and the context text to obtain an insurance segment category label; insurance liability association identification is performed on the entity to be disambiguated and the context text to obtain an insurance liability label; insurance user attribute association identification is performed on the entity to be disambiguated and the context text to obtain an insurance user attribute feature vector; and insurance business features are obtained based on the insurance segment category label, insurance liability label, and insurance user attribute feature vector.

[0119] In one embodiment, the fusion processing module 404 is used for:

[0120] The embedding vectors of the candidate entity in the insurance domain knowledge base, the insurance business features, and the context text are weighted and summed; based on the weighted summation result, the fusion vector corresponding to the candidate entity is obtained.

[0121] In one embodiment, the disambiguation result determination module 405 is used for:

[0122] The fusion vector and the entity to be disambiguated are input into a pre-constructed matching degree evaluation model; the matching degree between the fusion vector and the entity to be disambiguated is obtained based on the output of the matching degree evaluation model; and the candidate entities in the candidate entity set are sorted according to the relative size of the matching degree.

[0123] The modules in the aforementioned entity disambiguation processing device in the insurance field can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0124] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the data involved in the aforementioned methods. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements an entity disambiguation processing method in the field of insurance.

[0125] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0126] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the various method embodiments described above.

[0127] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the various method embodiments described above.

[0128] In one embodiment, a computer program product is provided having a computer program stored thereon, the computer program being executed by a processor of the steps described in the various method embodiments above.

[0129] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0130] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0131] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0132] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for entity disambiguation processing in the field of insurance, characterized in that, The method includes: Retrieve the entity to be disambiguated from the insurance document; In the insurance domain knowledge base, determine a set of candidate entities that match the entity to be disambiguated; In the insurance text, the context text of the entity to be disambiguated is determined, and insurance business features are extracted based on the entity to be disambiguated and the context text to obtain insurance business features; For each candidate entity in the candidate entity set, the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features and the context text are fused to obtain the fused vector corresponding to the candidate entity; Based on the matching degree between the fusion vector and the entity to be disambiguated, the candidate entities in the candidate entity set are sorted to determine the target candidate entity, so as to obtain the disambiguation result.

2. The method according to claim 1, characterized in that, Obtain the entities to be disambiguated in the insurance text, including: The insurance text is input into a pre-built insurance entity recognition model; the insurance entity recognition model is obtained by adjusting a natural language processing model based on an insurance dataset; Based on the results output by the insurance entity identification model, the entity to be disambiguated is obtained.

3. The method according to claim 1, characterized in that, A set of candidate entities matching the entity to be disambiguated is determined in the insurance domain knowledge base, including: When the standard name corresponding to one of the entities in the insurance domain knowledge base matches the entity to be disambiguated, the entity is determined as a candidate entity that matches the entity to be disambiguated. When the distance required for the entity to be disambiguated to be edited into one of the entities in the insurance domain knowledge base is less than a set distance threshold, the entity is determined as a candidate entity that matches the entity to be disambiguated. When the encoding information of the entity to be disambiguated matches the encoding information of one of the entities in the insurance domain knowledge base, the entity is determined as a candidate entity that matches the entity to be disambiguated. The candidate entity set is obtained based on the candidate entities that match the entity to be disambiguated.

4. The method according to claim 1, characterized in that, Based on the entity to be disambiguated and the context text, insurance business features are extracted to obtain insurance business features, including: The insurance segment type is identified by performing insurance segment type identification on the entity to be disambiguated and the context text to obtain insurance segment category labels; The entity to be disambiguated and the context text are used to identify insurance liability associations to obtain insurance liability tags. The insurance user attribute association recognition is performed on the entity to be disambiguated and the context text to obtain the insurance user attribute feature vector; Based on the insurance segment category tags, insurance liability tags, and insurance user attribute feature vectors, the insurance business characteristics are obtained.

5. The method according to claim 1, characterized in that, The embedding vectors of the candidate entities in the insurance domain knowledge base, the insurance business features, and the context text are fused to obtain the fused vectors corresponding to the candidate entities, including: The embedding vectors of the candidate entities in the insurance domain knowledge base, the insurance business features, and the embedding vectors of the context text are weighted and summed. Based on the weighted summation result, the fusion vector corresponding to the candidate entity is obtained.

6. The method according to any one of claims 1 to 4, characterized in that, Based on the matching degree between the fusion vector and the entity to be disambiguated, the candidate entities in the candidate entity set are sorted, including: The fusion vector and the entity to be disambiguated are input into a pre-constructed matching evaluation model; Based on the results output by the matching degree evaluation model, the matching degree between the fusion vector and the entity to be disambiguated is obtained; The candidate entities in the candidate entity set are sorted according to the relative size of the matching degree.

7. A physical disambiguation processing device in the field of insurance, characterized in that, The device includes: The entity identification module is used to obtain the entities to be disambiguated in the insurance text; The candidate entity recall module is used to determine a set of candidate entities that match the entity to be disambiguated in the insurance domain knowledge base; The business feature extraction module is used to determine the context text of the entity to be disambiguated in the insurance text, and to extract insurance business features based on the entity to be disambiguated and the context text to obtain insurance business features. The fusion processing module is used to fuse the embedding vector of the candidate entity in the insurance domain knowledge base, the embedding vector of the insurance business features and the embedding vector of the context text for each candidate entity in the candidate entity set, so as to obtain the fusion vector corresponding to the candidate entity. The disambiguation result determination module is used to sort the candidate entities in the candidate entity set according to the matching degree between the fusion vector and the entity to be disambiguated, and determine the target candidate entity to obtain the disambiguation result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

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

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

Images