Intelligent Outbound Call Knowledge Graph Construction Method and System Based on Dialogue Scenarios

Abstract

The present invention provides a method and system for constructing an intelligent outbound call knowledge graph based on a dialogue scenario, which relates to the technical field of voice knowledge graphs. In the present invention, the dialogue audio is real-time converted into text through a speech recognition API, and a syntactic tree is constructed by using a word segmentation tool and syntactic analysis to identify the part of speech and dependency relationship. Then, the syntactic annotation sequence is input into a BERT model for entity recognition to generate an entity set. A rule set is established to infer new relationships, and a graph theory algorithm is used to calculate the entity influence. Finally, the dependency relationship is mapped into the entity set, and a hierarchical index of each entity is generated according to the hierarchical relationship and influence index of the entities, and a multi-level knowledge graph is constructed through hierarchical mapping.

Description

Technical Field

[0001] The present invention relates to the technical field of speech knowledge graphs, and particularly to a method and system for constructing an intelligent outbound call knowledge graph based on a dialogue scenario. Background Art

[0002] In an intelligent outbound call system, the real-time processing and information extraction of voice conversations are the keys to achieving automation and intelligence. Traditional outbound call systems often rely on preset scripts and limited keyword matching, lacking flexibility and in-depth understanding of the conversation content. This limitation makes it difficult for the system to dynamically adapt to different conversation scenarios, unable to effectively extract valuable information from the conversation and update the knowledge base. In addition, traditional methods are often inefficient in processing complex relationships and multi-level information, and it is difficult to construct a knowledge graph that can comprehensively reflect the conversation content.

[0003] In the prior art, the published patent with the publication number CN113051405B discloses a method and device for constructing an intelligent outbound call knowledge graph based on a dialogue scenario: obtaining a script file related to the outbound call process, where the script file includes multiple dialogue topics and multiple dialogue contents corresponding to the multiple dialogue topics, and the multiple dialogue contents include dialogue contents related to the calling party and dialogue contents related to the called party; determining the connection relationship between the multiple dialogue topics according to the script file; determining the multiple dialogue contents according to the script file; and constructing a knowledge graph related to the outbound call process according to the determined connection relationship between the multiple dialogue topics and the multiple dialogue contents.

[0004] The main problems of the above method are: based on a static script file, the adaptability to the dialogue scenario is limited, and it is difficult to update or adjust the knowledge graph in a timely manner to reflect new information or changes in the conversation, resulting in insufficient applicability and flexibility of the knowledge graph, and determining the connection relationship between dialogue topics depends on manual judgment, and inaccurate judgment will directly affect the rationality of the knowledge graph; the system based on a preset script cannot fully consider the personalized needs and characteristics of each customer, affecting the customer experience.

[0005] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure, so it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for constructing an intelligent outbound call knowledge graph based on a dialogue scenario to solve the problems raised in the above background art.

[0007] To achieve the above purpose, the present invention provides the following technical solutions:

[0008] A method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario, the specific steps include:

[0009] Step 1: Real-time convert the dialogue audio into dialogue text through a speech recognition API, decompose the dialogue text into words by a word segmentation tool, annotate the part-of-speech of the words, generate a part-of-speech and syntactic annotation sequence of the dialogue text, generate a dependency relationship set based on the dialogue text with annotated part-of-speech, and construct a syntactic tree;

[0010] Step 2: Input the part-of-speech and syntactic annotation sequence of the dialogue text into a BERT model to generate context embedding vectors, and perform entity recognition on the words annotated in the dialogue text to generate a set of recognized entities;

[0011] Step 3: Construct a rule set, input the set of entities into the rule set to infer new relationships, and generate the influence of each entity based on graph theory algorithms;

[0012] Step 4: Map the dependency relationships in the dialogue text to the set of entities to generate the hierarchical relationship of the set of entities, generate a hierarchical index for each entity based on the upper and lower levels of the set of entities and the influence size of the entities, generate a set of hierarchical indexes based on the hierarchical indexes, construct a hierarchical mapping, and generate a multi-level knowledge graph.

[0013] Furthermore, the principle for constructing the syntactic tree is:

[0014] Mark a part-of-speech tag for each word according to the part-of-speech of the word. In the syntactic parser tool, identify the syntactic components of each word in the dialogue text, represent the syntactic relationship between each word and other words in the sentence, extract the dependency relationships between words, including subject-predicate relationship, verb-object relationship, modification relationship and preposition relationship. After confirming the dependency relationships, use the dependent word as the upper layer, and gradually supplement words in the parser to generate a syntactic tree.

[0015] Furthermore, the generated set of entities is expressed as follows:

[0016]

[0017] Generate a context embedding vector score for each word in the dialogue text through the BERT model. If the score of a word in the dialogue text is greater than the entity recognition threshold, integrate this word as an entity into the set of entities;

[0018] Among them, represents the set of entities, represents the i-th entity, represents the index of the entity, and I represents the number of entities.

[0019] Furthermore, the principle for constructing the rule set is:

[0020] Clarify the main purposes of intelligent outbound calls, including sales conversion, customer feedback collection, and service reminders, and set priorities from high to low in order.

[0021] Identify the key nodes in the conversation, including the opening greeting, need confirmation, and ending, and identify the possible intentions of the customer at each key node, including asking for details, expressing interest, and rejecting.

[0022] At each interaction node, determine the response instructions based on the possible intentions of the customer and formulate an "if - then" rule set.

[0023] Furthermore, the principle for generating the influence of each entity is as follows:

[0024]

[0025] Among them, represents the influence of entity ; represents the th entity; represents the damping coefficient; represents the set of other entities pointing to ; represents the set in which the entity pointing to is located; represents 's influence, and the initial value is the reciprocal of the number of entities; represents the number of entities pointing to other entities.

[0026] Furthermore, the principle for generating the hierarchical index is as follows:

[0027]

[0028]

[0029] Among them, represents the average value of entity influence; represents the total number of entities; represents the level of the i - th entity, and , M represents the total number of levels; represents rounding up the calculation result; represents the hierarchical index of the i - th entity.

[0030] Furthermore, the principle for constructing the hierarchical mapping is as follows:

[0031]

[0032] Among them, Indicates that the hierarchical index is of the mapping set, indicating the value of the hierarchical index, indicating the entities selected from the entity set, indicating the relationships selected from the dependency relationship set, Indicates that the hierarchical index is of the entity set, Indicates that the hierarchical index is the dependency relationship set corresponding to the entity;

[0033] All entities and dependency relationships are assigned to different levels according to different hierarchical indices. A subgraph is created in each level, containing all the entities and dependency relationships in that level. All subgraphs are spliced from high to low according to the size of the hierarchical index to generate a multi-level knowledge graph.

[0034] The present invention also provides an intelligent outbound call knowledge graph construction system based on a dialogue scenario. The system is used to implement the above-mentioned intelligent outbound call knowledge graph construction method based on a dialogue scenario, and specifically includes:

[0035] A speech recognition module, which is used to convert the dialogue audio into dialogue text in real time through a speech recognition API, decompose the dialogue text into words through a word segmentation tool, annotate the part-of-speech of the words, generate a part-of-speech and syntactic annotation sequence of the dialogue text, generate a dependency relationship set based on the dialogue text with annotated part-of-speech, and construct a syntactic tree;

[0036] An entity recognition module, which is used to input the part-of-speech and syntactic annotation sequence of the dialogue text into a BERT model to generate context embedding vectors, and perform entity recognition on the words annotated in the dialogue text to generate a set of recognized entities;

[0037] A rule construction module, which is used to construct a rule set, input the entity set into the rule set to infer new relationships, and generate the influence of each entity based on graph theory algorithms;

[0038] A knowledge graph construction module, which is used to map the dependency relationships in the dialogue text to the entity set to generate the upper and lower hierarchical relationships of the entity set, generate the hierarchical index of each entity based on the upper and lower levels of the entity set and the influence of the entity, generate a hierarchical index set based on the hierarchical index, construct a hierarchical mapping, and generate a multi-level knowledge graph.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] The present invention converts dialogue audio into dialogue text, ensuring the integrity and speech clarity of dialogue information during transmission. The constructed syntactic tree and dependency relationships provide important context for subsequent entity recognition and relationship reasoning, thereby improving the update efficiency of the knowledge base and the rationality of the hierarchical structure of the knowledge graph. By leveraging the powerful context understanding ability of the BERT model and deep learning techniques, in-depth semantic analysis of dialogue text is achieved. Through generating context embedding vectors, complex semantic relationships in dialogue text can be effectively understood, enhancing the accuracy of entity recognition.

[0041] The present invention also combines rule-based reasoning and graph theory algorithms to automatically infer potential new relationships from the recognized entity set and calculate the influence of each entity. Automatic reasoning and influence analysis enable the system to autonomously learn and optimize, enhancing the decision-making ability of the intelligent outbound calling system and improving the intelligence and practicality of the knowledge graph. Mapping the dependency relationships in dialogue text to the entity set can clearly construct the hierarchical relationships of entities and comprehensively consider the influence of entities, thereby creating a multi-level knowledge graph. By combining the hierarchical relationships and influence indices of entities, the importance of entities in the entire system can be more comprehensively evaluated, improving the practicality of the knowledge graph. Based on the dynamic adjustment ability of the hierarchical index, the knowledge graph can be quickly updated when information or relationships change, maintaining its timeliness, enabling the system to capture in-depth information in complex dialogues and providing strong support for further business decisions. Brief Description of the Drawings

[0042] Figure 1 It is a schematic flowchart of the method according to an embodiment of the present invention;

[0043] Figure 2 It is a schematic diagram of the system module according to an embodiment of the present invention. Detailed Embodiments

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with specific embodiments.

[0045] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Words such as "connected" or "linked" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Upper", "lower", "left", "right", etc. are only used to represent relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0046] Embodiment:

[0047] Please refer to Figure 1 , the present invention provides a technical solution:

[0048] A method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario, the specific steps include:

[0049] Step 1: Real-time convert the dialogue audio into dialogue text through a speech recognition API, decompose the dialogue text into words through a word segmentation tool, and mark the part-of-speech of the words to generate a part-of-speech and syntactic annotation sequence of the dialogue text. Based on the dialogue text with marked part-of-speech, generate a dependency relationship set and construct a syntactic tree.

[0050] In this embodiment, the principle for constructing the syntactic tree is:

[0051] Mark a part-of-speech tag for each word according to the part-of-speech of the word. In the syntactic parser tool, identify the syntactic components of each word in the dialogue text, represent the syntactic relationship between each word and other words in the sentence, extract the dependency relationships between words, including subject-predicate relationship, verb-object relationship, modification relationship and preposition relationship. After confirming the dependency relationships, use the dependent words as the upper layer and gradually supplement words in the parser to generate a syntactic tree.

[0052] Real-time convert the dialogue audio into text through Google Speech-to-Text, decompose the text into independent words through the natural language processing tool SpaCy, assign a corresponding grammatical category to each word, such as noun, verb, adjective, etc., generate a dependency graph, the nodes of the graph are words, and the edges of the graph represent the dependency relationships between words. Convert the dependency graph into a tree structure, which is the syntactic tree. The root node of the syntactic tree is the core verb or predicate word of the sentence, and other words are connected as child nodes in sequence to form a tree-like hierarchy.

[0053] Step 2: Input the part-of-speech and syntactic annotation sequence of the dialogue text into the BERT model to generate context embedding vectors, and perform entity recognition on the annotated words in the dialogue text to generate the set of recognized entities;

[0054] The set of entities generated in this embodiment is expressed as follows:

[0055]

[0056] Generate the context embedding vector score for each word in the dialogue text through the BERT model. If the score of a word in the dialogue text is greater than the entity recognition threshold, integrate this word as an entity into the set of entities;

[0057] Among them, represents the set of entities, represents the i-th entity, represents the index of the entity, and I represents the number of entities.

[0058] Only when the context embedding vector score is greater than the entity recognition threshold will the word be recognized as an entity; the principle for generating the entity recognition threshold is as follows:

[0059] Calculate the mean and standard deviation of the context embedding vector scores for each word in the dialogue text:

[0060]

[0061]

[0062]

[0063] Among them, represents the mean of the context embedding vector scores, represents the index of the context embedding vector scores, represents the number of the context embedding vector scores, represents the -th context embedding vector score, represents the standard deviation of the context embedding vector scores, represents the entity recognition threshold.

[0064] Step 3: Construct a rule set, input the set of entities into the rule set to infer new relationships, and generate the influence of each entity based on the graph theory algorithm;

[0065] In this embodiment, the principle for constructing the rule set is as follows:

[0066] Clarify the main purposes of intelligent outbound calls, including sales conversion, customer feedback collection, and service reminders, and set priorities from high to low in order;

[0067] Identify the key nodes in the conversation, including the opening greeting, demand confirmation, and ending, and identify the possible intentions of the customer at each key node, including asking for details, expressing interest, and rejecting;

[0068] At each interaction node, determine the response instructions based on the possible intentions of the customer and formulate an "if - then" rule set.

[0069] Define clear goals, identify the types of outbound calls, including sales, surveys, reminders, etc., distinguish the goals into short - term goals and long - term goals, and set evaluation criteria respectively, such as sales volume, survey completion rate, customer satisfaction, etc. Assign weights to different goals to guide resource allocation; Integrate multi - channel data including call records, CRM systems, and customer feedback to ensure data consistency; Create a customer interaction flow chart, identify the key nodes in the conversation, including the opening greeting, demand confirmation, ending, etc. The main requirements include product introduction and question answering. Identify customer reactions and potential needs at each node, establish a customer intention library covering common questions and needs based on historical data, and based on the customer intention library, establish a rule library of "if + condition, then + action", and set dynamic priorities for the rule set according to the completion of the evaluation criteria of the business goals. The dialogue categories with higher dynamic priorities are executed first. The formula for setting dynamic priorities is:

[0070]

[0071] Among them, represents the dynamic priority of the goal, represents the completion standard of the goal, represents the actual completion of the goal;

[0072] The more the actual completion volume of the goal, the lower the dynamic priority, and the later it is executed. Other goals with less actual completion are executed first.

[0073] The principle for generating the influence of each entity is:

[0074]

[0075] Among them, represents the influence of entity , represents the th entity, represents the damping coefficient, represents pointing to the set of other entities, represents the set The entity pointed to in indicates the influence of and the initial value is the reciprocal of the number of entities. indicates the number of entities pointed to by other entities.

[0076] In graph theory algorithms, each entity corresponds to a node, reflecting the entity node in the importance within the entire link structure, measuring the degree to which the node is cited or pointed to by other nodes and the importance of those citing or pointing nodes themselves. If a node is pointed to by multiple important nodes, the pointed-to node is also an important node with relatively high influence. The damping factor is used to simulate the probability of randomly jumping to other entity nodes;

[0077] Step 4: Map the dependency relationships in the dialogue text to the entity set, generate the hierarchical relationship of the entity set, generate the hierarchical index for each entity based on the hierarchical relationship of the entity set and the influence of the entity, and generate a hierarchical set based on the hierarchical index, construct a hierarchical mapping, and generate a multi-level knowledge graph.

[0078] The principle for generating the hierarchical index is as follows:

[0079]

[0080]

[0081] Among them, represents the average value of entity influence, represents the total number of entities, represents the level of the i-th entity, and , M represents the total number of levels, represents rounding up the calculation result, represents the hierarchical index of the i-th entity.

[0082] The hierarchical index comprehensively reflects the level and importance of the entity. The higher the level of the entity, the higher the influence, indicating that the entity is more important, and the corresponding hierarchical index is higher. The hierarchical index is proportional to the level and influence of the entity. The entity at the top of the hierarchical relationship corresponds to the level M, and the entity at the bottom corresponds to the level 1.

[0083] The principle for constructing the hierarchical mapping is as follows:

[0084]

[0085] Among them, Indicates the mapping set with a hierarchical index of , Indicates the value range of the hierarchical index, Indicates the entities selected from the entity set, Indicates the relationships selected from the dependency relationship set, Indicates the entity set with a hierarchical index of , Indicates the entity set with a hierarchical index of and the corresponding dependency relationship set of the entity.

[0086] In this embodiment, the hierarchy represents the upper and lower relationships determined based on the dependency relationships between entities, and the level represents the upper and lower relationships determined according to the hierarchical index of the entities;

[0087] The purpose of constructing the hierarchical mapping is to assign entities with the same hierarchical index to the same level. The higher the hierarchical index of an entity, the higher its corresponding influence, and the more important the entity. Entities with higher hierarchical indices are placed at higher levels.

[0088] All entities and dependency relationships are assigned to different levels according to different hierarchical indices. A subgraph is created in each level, which contains all the entities and dependency relationships in that level. All subgraphs are spliced from high to low according to the size of the hierarchical index to generate a multi-level graph spectrum.

[0089] Please refer to Figure 2 , the present invention also provides an intelligent outbound call knowledge graph construction system based on a dialogue scenario. The system is used to implement the above-mentioned intelligent outbound call knowledge graph construction method based on a dialogue scenario, and specifically includes:

[0090] A speech recognition module, which is used to convert the dialogue audio into dialogue text in real time through a speech recognition API, decompose the dialogue text into words through a word segmentation tool, annotate the part-of-speech of the words, generate a part-of-speech and syntactic annotation sequence of the dialogue text, generate a dependency relationship set based on the dialogue text with annotated part-of-speech, and construct a syntactic tree;

[0091] An entity recognition module, which is used to input the part-of-speech and syntactic annotation sequence of the dialogue text into a BERT model to generate context embedding vectors, and perform entity recognition on the words annotated in the dialogue text to generate a set of recognized entities;

[0092] A rule construction module, which is used to construct a rule set, input the entity set into the rule set to infer new relationships, and generate the influence of each entity based on graph theory algorithms;

[0093] The graph construction module is used to map the dependency relationships in the dialogue text into an entity set, generate the hierarchical relationships of the entity set, generate the hierarchical index for each entity based on the hierarchical levels of the entity set and the influence magnitude of the entities, and generate a set of hierarchical indices based on the hierarchical indices, construct a hierarchical mapping, and generate a multi-level knowledge graph.

[0094] All the above formulas are calculated by taking the numerical values after dimensionless processing. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation.

[0095] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0096] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0097] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered within the protection scope of this application.

Claims

1. A method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario, characterized in that The specific steps include: Step 1: Real-time convert the conversation audio into conversation text through a speech recognition API, decompose the conversation text into words by a word segmentation tool, label the part-of-speech of the words, generate a part-of-speech and syntactic annotation sequence of the conversation text, generate a set of dependency relationships based on the conversation text with labeled part-of-speech, and construct a syntactic tree; Step 2: Input the part-of-speech and syntactic annotation sequence of the conversation text into the BERT model to generate context embedding vectors, and perform entity recognition on the words labeled in the conversation text to generate a set of recognized entities; Step 3: Construct a rule set, input the set of entities into the rule set to infer new relationships, and generate the influence of each entity based on graph theory algorithms; Step 4: Map the dependency relationships in the conversation text to the set of entities to generate the hierarchical relationship of the set of entities, generate a hierarchical index for each entity based on the upper and lower levels of the set of entities and the influence size of the entities, generate a set of hierarchical indexes based on the hierarchical indexes, construct a hierarchical mapping, and generate a multi-level knowledge graph; The principle for generating the hierarchical index is: Among them, represents the average value of entity influence, I represents the total number of entities, m i represents the level of the i-th entity, and m i ∈[1, M], M represents the total number of levels, represents rounding up the calculation result, F i represents the level index of the i-th entity; The principle for constructing the hierarchical mapping is: N(l) = {(e,r)|e∈e(l),r∈R(l)} Among them, N(l) represents the mapping set with a hierarchical index of l, l represents the value of the hierarchical index, e represents the entity selected from the set of entities, r represents the relationship selected from the set of dependency relationships, E(l) represents the set of entities with a hierarchical index of l, and R(l) represents the set of dependency relationships corresponding to the entities with a hierarchical index of l; All entities and dependency relationships are assigned to different levels according to different hierarchical indexes. A subgraph is created in each level, which contains all the entities and dependency relationships of that level. All subgraphs are spliced from high to low according to the size of the hierarchical indexes to generate a multi-level graph.

2. The method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario according to claim 1, wherein: The principle for constructing the syntactic tree in the above Step 1 is: Mark a part-of-speech tag for each word according to the part-of-speech of the word. In the syntactic parser tool, identify the syntactic components of each word in the conversation text, represent the syntactic relationship between each word and other words in the sentence, extract the dependency relationships between words, including subject-predicate relationship, verb-object relationship, modification relationship and preposition relationship. After confirming the dependency relationships, use the dependent word as the upper layer, and gradually supplement words in the parser to generate a syntactic tree.

3. The method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario according to claim 1, characterized in that: The set of entities generated in the above Step 2 is described as follows: E = {e i | i ∈ [1, I]} Generate a context embedding vector score for each word in the conversation text through the BERT model. If the score of a word in the conversation text is greater than the entity recognition threshold, integrate this word as an entity into the set of entities; Among them, E represents the entity set, and e i represents the i-th entity, i represents the index of the entity, and I represents the number of entities.

4. A method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario according to claim 1, characterized in that: The principle for constructing the rule set in the above Step 3 is: Clarify the main purposes of intelligent outbound calls, including sales conversion, customer feedback collection, service reminder, and set priorities from high to low in order; Identify the key nodes in the conversation, including opening greetings, demand confirmation, ending, and identify the possible intentions of the customer at each key node, including asking for details, expressing interest and rejection; At each interaction node, determine the reply instructions based on the possible intentions of the customer, and formulate an "if-then" rule set.

5. A method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario according to claim 1, characterized in that: The principle for generating the influence of each entity in step 3 is as follows: Among them, P(e i ) represents the influence of entity e i , e i represents the i-th entity, d represents the damping coefficient, IN(e i ) represents the set of other entities pointing to e i , e n represents the entity in the set IN(e i ) that points to e i , P(e n ) represents the influence of e n , and its initial value is the reciprocal of the number of entities. L(e n ) represents the number of entities that e n points to other entities.

6. An intelligent outbound call knowledge graph construction system based on a dialogue scenario, characterized in that: The system is used to implement the method for constructing an intelligent outbound call knowledge graph based on a dialogue scenario described in any one of claims 1-5, specifically including: A speech recognition module, which is used to convert the dialogue audio into dialogue text in real time through a speech recognition API, decompose the dialogue text into words through a word segmentation tool, annotate the part-of-speech of the words, generate a part-of-speech syntactic annotation sequence of the dialogue text, generate a dependency relationship set based on the dialogue text with annotated part-of-speech, and construct a syntactic tree; An entity recognition module, which is used to input the part-of-speech syntactic annotation sequence of the dialogue text into a BERT model to generate context embedding vectors, and perform entity recognition on the words annotated in the dialogue text to generate a set of recognized entities; A rule construction module, which is used to construct a rule set, input the entity set into the rule set to infer new relationships, and generate the influence of each entity based on graph theory algorithms; A knowledge graph construction module, which is used to map the dependency relationships in the dialogue text to the entity set to generate the hierarchical relationship of the entity set, generate a hierarchical index for each entity based on the upper and lower levels of the entity set and the influence size of the entity, generate a set of hierarchical indexes based on the hierarchical indexes, construct a hierarchical mapping, and generate a multi-level knowledge graph.

Images