Dialogue control method, system storage medium and server based on subject object tracking and cognitive reasoning

Abstract

The present invention discloses a dialogue control method, a system storage medium and a server based on subject object tracking and cognitive reasoning, wherein the method comprises the following steps: constructing an event object database; constructing a matter map corresponding to the event object; comparing the dialogue text data and the event object to identify the event object; tracking the event object in real time; constructing an event object database, and constructing a matter map corresponding to the event object; when the user inputs a dialogue text, the processor controls the recognition module to compare the word vector value of the dialogue text data and the word vector value of the event object to identify the event object, and the processor tracks the event object in real time by controlling the tracking module, and calls the matter map corresponding to the event object, and generates reply content by text analysis. Through the above-mentioned setting, the present invention endows the system with the ability of tracking and cognitive reasoning of the subject object, and improves the accuracy and logical judgment ability of the robot in identifying the conversation subject.

Description

Technical Field

[0001] The present invention relates to the field of customer service centers, and in particular to a dialogue control method based on subject object tracking and cognitive reasoning, a system storage medium and a server. Background Art

[0002] Artificial intelligence technology plays an important role in today's customer service field. Conversational robots based on artificial intelligence technology are a new way to build interactions between enterprises and customers. In the service process of conversational robots, the robots need to understand the intentions in the customer's words, match them to the corresponding business dialogue process or knowledge points, and complete the response to the user.

[0003] In the process of dialogue interaction, due to the colloquial expression of customers, subjects are often missing in multiple rounds of dialogue. It is difficult to correctly match knowledge points using deep learning text similarity calculation models or classification models when subjects are missing. In addition, since the business dialogue process is a time-series node structure constructed manually, when customers jump out of the preset dialogue process sequence or require robots to make logical judgments, an endless loop of logical missing often occurs, making it difficult for this traditional dialogue process to proceed effectively.

[0004] For example, in the tax consultation scenario, unlike the fixed question-and-answer mode of simple business, when answering tax-related questions, the robot will encounter the following difficulties: if the value of any object such as taxpayer category, tax type, tax-related goods changes, it should match the answer that is suitable for the consultant's own situation. However, there are many restrictions on knowledge points in this scenario, and some knowledge points also have multiple layers of inclusion relationships at the same time, and users often express incomplete information when consulting. For example, a taxpayer asks: "Can the input tax of my handling fee be deducted?" The robot recognizes it as a consultation intention of "input tax deduction", but necessary conditions such as taxpayer, invoice type, project / service, etc. are missing, and traditional robots cannot obtain complete information.

[0005] Even with complete information, traditional robots can only find the relationship between "input tax" and "tax deduction" using the knowledge graph method. However, due to the lack of reasoning ability, robots cannot determine whether this specific tax can be deducted, and consultants often return empty-handed. Using business dialogue processes is quite a match, and you can clarify with users step by step and sort out the restrictions, but there are also shortcomings: because these businesses need to preset various processes and conditions to make dialogue processes, the final result is an overly large process system, which is very bloated and also faces the problem of high operating costs in the later stage.

[0006] In order to improve the robot's accuracy in identifying conversation topics and its logical judgment ability, the present application provides a dialogue control system based on subject object tracking and cognitive reasoning. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a dialogue control method based on subject object tracking and cognitive reasoning, a system storage medium and a server.

[0008] In a first aspect, a dialogue control method based on subject object tracking and cognitive reasoning comprises the following steps:

[0009] Build event object database through business rule word library and product word library;

[0010] Construct a matter-of-fact graph corresponding to the event object through the first method;

[0011] The user inputs the conversation text data;

[0012] comparing the conversation text data and the event object to identify the event object;

[0013] Real-time tracking of event objects;

[0014] Construct an event object database through the first construction module, and construct a matter graph corresponding to the event object through the second construction module;

[0015] When the user inputs a conversation text, the processor controls the recognition module to compare the word vector value of the conversation text data with the word vector value of the event object to identify the event object. The processor controls the tracking module to track the event object in real time, calls the event graph corresponding to the event object, and generates reply content through text analysis.

[0016] In the second aspect, a dialogue control system based on subject object tracking and cognitive reasoning includes:

[0017] The first construction module constructs an event object database through a business rule vocabulary and a product vocabulary;

[0018] A second construction module, which constructs a matter graph corresponding to the event object by using the first method;

[0019] An input module, which is used for users to input dialogue text data;

[0020] A recognition module that recognizes an event object by comparing the conversation text data with the event object;

[0021] A tracking module, which is used to track event objects in real time;

[0022] A processor, which constructs an event object database through a first construction module, and constructs a matter graph corresponding to the event object through a second construction module;

[0023] When the user inputs a conversation text, the processor controls the recognition module to compare the word vector value of the conversation text data with the word vector value of the event object to identify the event object. The processor controls the tracking module to track the event object in real time, calls the event graph corresponding to the event object, and generates reply content through text analysis.

[0024] Further: Also includes

[0025] A first database pre-stored with knowledge base text data;

[0026] A business rule word library, which is used to store the first key word and the second key word;

[0027] A second database pre-stored with relational database information;

[0028] Product word library, which is used to store the name information corresponding to the primary key;

[0029] The step of "a first construction module, which constructs an event object database through a business rule word library and a product word library" is: the processor uses a new word discovery algorithm of a text processing tool to calculate the information entropy and mutual information of the knowledge base text data in the first database to mine unregistered words, and add them to the first database;

[0030] In the top-level node of the knowledge base text data in the first database, the processor uses a first extraction method to extract a first key word and adds it to the business rule vocabulary;

[0031] At the end node of the knowledge base text data in the first database, the processor uses a second extraction method to extract a second key word and adds it to the business rule word library;

[0032] The processor extracts relational database information from the second database using an information extraction method, extracts name information corresponding to the primary key, and adds the name information to the product word library;

[0033] The processor controls the first construction module to merge the business rule vocabulary and the product vocabulary into an event object database.

[0034] Further: the first database also pre-stores a knowledge graph;

[0035] The first method is that the processor can refer to the hierarchical relationship of entity data in the knowledge graph to construct a logic graph of hierarchical reasoning relationship.

[0036] Further: the second database also pre-stores unstructured text;

[0037] The first method is that the processor can identify the relational database information and unstructured text in the second database through a relational learning model, and construct a logical graph of constituent reasoning, reverse reasoning, and concurrent reasoning relationships.

[0038] Further: the step of “the identification module, which identifies the event object by comparing the conversation text data and the event object” is:

[0039] Based on the event object database, the processor calculates the first word vector value of each event object data in the event object database through a vector calculation model;

[0040] The processor is capable of converting conversation text data into an input event object, and calculating a second word vector value corresponding to the input event object through a vector calculation model. The processor uses a vector similarity calculation method to determine whether the similarity between the first word vector value and the second word vector value is greater than a first preset threshold. If so, the input event object is determined to be the event object with a similarity greater than the first preset threshold, and is stored in an event object database. If not, the event object is cached through a regular rule method.

[0041] Further: the step of “the tracking module, which is used to track the event object in real time” is: the processor compares the conversation text data and the event object in real time through the recognition module to recognize the event object;

[0042] The processor can obtain first time data of the dialogue text data input by the user through the input module, and convert the event object corresponding to the dialogue text data corresponding to the first time data into a priority event object.

[0043] Further: Also includes

[0044] A receiving module, which is used for users to receive reply content data;

[0045] The step of "calling the cause-and-effect graph corresponding to the event object and generating reply content through text analysis" is: calling the cause-and-effect graph corresponding to the priority event object, combining the logical operation method, outputting the reply content data and sending it to the receiving module.

[0046] In a third aspect, a computer-readable storage medium stores a computer program, which implements the steps of the above method when executed by a processor.

[0047] In a fourth aspect, a server comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method when executing the computer program.

[0048] The difference between the present invention and the prior art lies in that the present invention endows the system with the ability to track and cognitively reason the subject object, thereby improving the robot's accuracy in identifying the conversation subject and its logical judgment ability. Subject object tracking is an identification capability that is completely built on industry data, and can accurately identify the event object of the conversation. It is highly controllable and more suitable for closed fields in professional industries.

[0049] The following is a further description of a dialogue control system based on subject object tracking and cognitive reasoning of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of a dialogue control system based on subject object tracking and cognitive reasoning.

[0051] Figure 2 It is a flowchart for constructing an event object database;

[0052] Figure 3 It is a flow chart of event object recognition;

[0053] Figure 4 It is a flow chart of event object tracking;

[0054] Figure 5 It is a flow chart for constructing a logical map. DETAILED DESCRIPTION

[0055] A dialogue control method based on subject object tracking and cognitive reasoning includes the following steps:

[0056] Build event object database through business rule word library and product word library;

[0057] Construct a matter-of-fact graph corresponding to the event object through the first method;

[0058] The user inputs conversation text data;

[0059] comparing the conversation text data and the event object to identify the event object;

[0060] Real-time tracking of event objects;

[0061] Construct an event object database through the first construction module, and construct a matter graph corresponding to the event object through the second construction module;

[0062] When the user inputs a conversation text, the processor controls the recognition module to compare the word vector value of the conversation text data with the word vector value of the event object to identify the event object. The processor controls the tracking module to track the event object in real time, calls the event graph corresponding to the event object, and generates reply content through text analysis.

[0063] like Figures 1 to 5 As shown, see Figure 1 The present invention provides a dialogue control system based on subject object tracking and cognitive reasoning, including

[0064] The first construction module constructs an event object database through a business rule vocabulary and a product vocabulary;

[0065] A second construction module, which constructs a matter graph corresponding to the event object by using the first method;

[0066] An input module, which is used for users to input dialogue text data;

[0067] A recognition module that recognizes an event object by comparing the conversation text data with the event object;

[0068] A tracking module, which is used to track event objects in real time;

[0069] A processor, which constructs an event object database through a first construction module, and constructs a matter graph corresponding to the event object through a second construction module;

[0070] When the user inputs a conversation text, the processor controls the recognition module to compare the word vector value of the conversation text data with the word vector value of the event object to identify the event object. The processor controls the tracking module to track the event object in real time, calls the event graph corresponding to the event object, and generates reply content through text analysis.

[0071] The present invention preliminarily constructs an event object database using a business rule vocabulary and a product vocabulary through a first construction module, and constructs a matter graph corresponding to the event object using a second construction module for backup. The conversation text data input by the user is recognized and tracked in real time through an identification module and a tracking module, and then the corresponding constructed matter graph is electrophoresed to give the system cognitive reasoning capabilities, and finally the user is replied through analysis of the conversation text data.

[0072] Through the above-mentioned configuration, the present invention endows the system with the ability to track and cognitively reason about subject objects, thereby improving the robot's accuracy in identifying conversation topics and its logical judgment ability.

[0073] The event object database stores a plurality of event object data, and the event object data can be understood as the user's consultation intention contained in the dialogue text data input when the user consults a question.

[0074] Among them, the first method can be: construction by industry experts, reference to knowledge graphs, use of relational databases, etc.

[0075] Among them, the said event graph is an event reasoning graph, which can enable the robot to have cognitive reasoning ability without relying on the traditional knowledge graph. The event graph is different from the knowledge graph in that it can only describe and store static knowledge data. The knowledge graph cannot solve the problems of "why", "what to do", "is it possible" and so on. These words represent the process of human thinking. We define the thinking process as the process of using "event reasoning". The so-called "event" is the reason of "events" and the path of thinking. This "event" is the event we need.

[0076] Among them, the logical relationships contained in the event map include causal relationship, conditional relationship, exclusive relationship, sequential relationship, superior-subordinate relationship, composition relationship and concurrent relationship. These relationships describe the logical knowledge between event subjects. Therefore, the event map has seven reasoning abilities, including causal reasoning, conditional reasoning, exclusive reasoning, sequential reasoning, superior-subordinate reasoning, composition reasoning and concurrent reasoning.

[0077] Wherein, the input module can be: a mobile phone, a computer or other text input devices.

[0078] For example, if the conversation text data input by the user is: "Tibet is a beautiful place, I want to go and see it", then the consultation intention may be the time and method of travel to Tibet, etc. By calling the Tibet travel event map, the following reply content can be deduced through text analysis: air ticket reservation, hotel reservation, train ticket reservation, etc.

[0079] For further explanation of the present invention, see Figure 1 , 2 , a first database, which pre-stores knowledge base text data;

[0080] A business rule word library, which is used to store the first key word and the second key word;

[0081] A second database pre-stored with relational database information;

[0082] Product word library, which is used to store the name information corresponding to the primary key;

[0083] The step of "a first construction module, which constructs an event object database through a business rule word library and a product word library" is: the processor uses a new word discovery algorithm of a text processing tool to calculate the information entropy and mutual information of the knowledge base text data in the first database to mine unregistered words, and add them to the first database;

[0084] In the top-level node of the knowledge base text data in the first database, the processor uses a first extraction method to extract a first key word and adds it to the business rule vocabulary;

[0085] At the end node of the knowledge base text data in the first database, the processor uses a second extraction method to extract a second key word and adds it to the business rule word library;

[0086] The processor extracts relational database information from the second database using an information extraction method, extracts name information corresponding to the primary key, and adds the name information to the product word library;

[0087] The processor controls the first construction module to merge the business rule vocabulary and the product vocabulary into an event object database.

[0088] The present invention realizes the construction of an event object database through the above-mentioned setting, which facilitates the identification and tracking of event objects in subsequent steps.

[0089] Among them, the top node is the node of the first element, which is the first node after the head node.

[0090] Among them, the end node is the last node in the linked list in the data structure, that is, the node that stores the last element, and the corresponding node is the head node, which is a node attached before the first node of the linked list.

[0091] The algorithm logic of the new word discovery algorithm is mainly divided into three steps: 1. Convert the corpus text into a string, then generate an n_gram dictionary, and count the frequency of each word. 2. Use the point-to-point mutual information to filter out candidate new words from the previous n_gram dictionary. 3. Filter out the final output new words from the candidate new words through left and right entropy, which is a prior art and will not be repeated here.

[0092] Among them, out of vocabulary (OOV) refers to words that did not appear during training but appeared during testing. In natural language processing or text processing, there is usually a vocabulary, which is either pre-loaded, self-defined, or extracted from the current dataset. Suppose there is another dataset later, and some words in this dataset are not in the existing vocabulary, then these words are said to be out of vocabulary, or OOV for short.

[0093] The first extraction method may use the Text Rank algorithm. The Text Rank algorithm is a graph-based sorting algorithm for text, which divides the text into several components (sentences), constructs a node connection graph, uses the similarity between sentences as the weight of the edge, calculates the Text Rank value of the sentence through a loop iteration, and finally extracts the high-ranking sentences to form a text summary.

[0094] The second extraction method may use the TF-IDF (term frequency–inverse document frequency) algorithm. The TF-IDF algorithm is a statistical method used to evaluate the importance of a word to a document set or one of the documents in a corpus. The importance of a word increases in direct proportion to the number of times it appears in a document, but at the same time decreases in inverse proportion to the frequency of its appearance in the corpus.

[0095] For further explanation of the present invention, see Figure 1 , 5 , the first database also pre-stores a knowledge graph;

[0096] The first method is that the processor can refer to the hierarchical relationship of entity data in the knowledge graph to construct a logic graph of hierarchical reasoning relationship.

[0097] The present invention constructs a logic diagram of upper and lower reasoning through the first method, which is convenient for calling and analyzing the dialogue text data.

[0098] The knowledge graph is a structured semantic knowledge base used to quickly describe concepts and their interrelationships in the physical world. It reduces the data granularity from the document level to the data level, aggregates a large amount of knowledge, and thus achieves rapid response and reasoning of knowledge. If "knowledge graph" is considered to represent a knowledge base in a broad sense, which is an ontology used to store knowledge, then "matter graph" is a "knowledge graph" that stores the logical relationship between things and things; if "knowledge graph" is considered to refer specifically to the entity-centered knowledge base built by Google, Baidu, etc. in a narrow sense to enhance the user search experience, then "matter graph" is a new type of common sense knowledge base that is parallel to "knowledge graph".

[0099] There are two types of hierarchical relationships: noun hierarchical relationships and verbal hierarchical relationships. For example, the events "food price increases" and "vegetable price increases" are noun hierarchical relationships; the events "killing" and "assassination" are verbal hierarchical relationships. It should be noted that hierarchical relationships are generally deterministic knowledge, so there is no need to analogize sequential or causal relationships. A constant between 0 and 1 is assigned to the hierarchical relationship to indicate its confidence.

[0100] Among them, the hierarchical reasoning describes the logic of events in the classification system; a certain event is a superior or subordinate event of another time; it is formalized as Up A down B.

[0101] Among them, the first method can also be: through industry experts based on their own rich knowledge accumulation, quickly build a business and product reasoning map, and quickly form a reasoning map of conditional reasoning, causal reasoning and sequential reasoning.

[0102] The present invention constructs a logic diagram of conditional reasoning, causal reasoning and sequential reasoning through the first method, which is convenient for calling and analyzing the dialogue text data.

[0103] Among them, conditional reasoning describes a condition-result relationship in the cognitive reasoning system, which is a kind of premise and result logic; it is formalized as if A then B; causal reasoning describes a cause-and-effect relationship in the cognitive reasoning system, where the previous event leads to the occurrence of the next event; it is formalized as cause A so B; sequential reasoning describes a temporal partial order relationship in the cognitive reasoning system, which is a kind of sequential action logic; it is formalized as first A then B.

[0104] For example, when the robot sells products over the phone, if the customer's feedback is that the purchase intention is not strong, the conditional reasoning relationship of the marketing reasoning map can be located. If it is identified that the product is expensive, a discount should be replied. If it is identified that the performance is not good, a high-performance model should be recommended as an alternative. When registering for the APP, if the customer fails to upload information, causal reasoning can be established. When the customer asks how to solve it, the robot calls the registered causal reasoning map, actively collects the reasons for the failure, and gives corresponding solutions based on the reasons for the failure. When the customer actively expresses that Tibet is a beautiful place and wants to go and see it, the intention can be identified as a travel time. Through the Tibet travel sequential reasoning map, many consumer behaviors can be launched, including air ticket reservations, hotel reservations, train ticket reservations, etc.

[0105] For further explanation of the present invention, see Figure 1 , 5 , the second database also pre-stores unstructured text;

[0106] The first method is that the processor can identify the relational database information and unstructured text in the second database through a relational learning model, and construct a logical graph of constituent reasoning, reverse reasoning, and concurrent reasoning relationships.

[0107] The present invention uses the first method to construct a logical diagram of composition reasoning, reverse reasoning, and concurrent reasoning relationships, which is convenient for calling and analyzing dialogue text data.

[0108] Among them, the relational learning model can adopt word embedding, predicate dictionary and other methods. Word embedding is a method of converting words in a text into digital vectors. In order to use standard machine learning algorithms to analyze them, these vectors converted into numbers need to be input in digital form. The predicate dictionary is the most basic formal system of mathematical logic, which is also called first-order logic. A proposition that can answer true or false can be analyzed not only into simple propositions, but also into individuals, quantifiers and predicates. An individual represents an object or element, a quantifier represents a quantity, and a predicate represents an attribute of an individual.

[0109] Among them, compositional reasoning describes the logic of the whole and the parts between events; it is a relational logic of inclusion and being included; it is formalized as A includes B; inversion reasoning often describes a mutually exclusive logic in the cognitive reasoning system, which is a true and false value logic; it is formalized as although A but B; concurrent reasoning describes a symbiotic relationship between events in time, which means that when one event occurs, another event must occur, formalized as A coincide B.

[0110] For example, when a customer asks whether the tax amount of the handling fee can be deducted, the component reasoning graph of the tax amount deduction is located, the support range of the deduction is found, and it is determined whether the handling fee is included in the component reasoning relationship; in banking business, the user handles business A and business B at the same time, and in fact, there is a mutually exclusive relationship between handling business A and handling business B. Then, the user's business knowledge information can be updated according to the order of handling the two businesses, and the knowledge of handling business B can be removed.

[0111] For further explanation of the present invention, see Figure 1 , 3 , the steps of “the identification module, which identifies the event object by comparing the conversation text data and the event object” are:

[0112] Based on the event object database, the processor calculates the first word vector value of each event object data in the event object database through a vector calculation model;

[0113] The processor is capable of converting conversation text data into an input event object, and calculating a second word vector value corresponding to the input event object through a vector calculation model. The processor uses a vector similarity calculation method to determine whether the similarity between the first word vector value and the second word vector value is greater than a first preset threshold. If so, the input event object is determined to be the event object with a similarity greater than the first preset threshold, and is stored in an event object database. If not, the event object is cached through a regular rule method.

[0114] The present invention first calculates the first word vector value of each event object data in the event object database, then calculates the second word vector value of the input event object corresponding to the conversation text data, and determines whether the similarity between the first word vector value and the second word vector value is greater than a first preset threshold. If so, it means that the input event object corresponding to the conversation text data input by the user is the event object compared above, and it is updated to the event object database.

[0115] The present invention achieves matching of the dialogue text data input by the user with the pre-stored event object through the above-mentioned method, so as to facilitate the subsequent real-time tracking of the event object and the retrieval of the event graph corresponding to the event.

[0116] The vector calculation model is a tool for document representation and similarity calculation, which is widely used not only in the search field but also in natural language processing, text mining and other fields. As a tool for representing documents, the vector space model regards each document as a vector composed of t-dimensional features. The definition of features can be adopted in different ways. The most common one is to use words as features, that is, to extract t keywords from a document, where each feature will calculate its weight according to a certain algorithm, and this t-dimensional feature vector with weights is used to represent this document.

[0117] The vector similarity calculation method may be: Euclidean distance, Manhattan distance, Chebyshev distance, Minkowski distance, etc. It calculates the distance between two vectors. The closer the distance between the two vectors, the more similar the two vectors are. Of course, angle cosine, inverse relationship, etc. may also be used.

[0118] Among them, the regular rule method is: regular expression, which describes a string matching pattern, which can be used to check whether a string contains a certain substring, replace the matching substring, or extract the substring that meets a certain condition from a string. The method of constructing a regular expression is the same as the method of creating a mathematical expression. That is, small expressions can be combined to create larger expressions using a variety of metacharacters and operators. The components of a regular expression can be a single character, a character set, a character range, a choice between characters, or any combination of all these components.

[0119] For further explanation of the present invention, see Figure 1 , 4 , the step of “the tracking module, which is used to track the event object in real time” is: the processor compares the conversation text data and the event object in real time through the recognition module to identify the event object;

[0120] The processor can obtain first time data of the dialogue text data input by the user through the input module, and convert the event object corresponding to the dialogue text data corresponding to the first time data into a priority event object.

[0121] The present invention identifies the event objects corresponding to the input conversation text data in real time, and increases the priority of the event objects at the most recent time, while the priorities of other event objects are relatively reduced, so that the event graph corresponding to the above event objects is first called in the subsequent steps, thereby achieving real-time tracking of the event objects corresponding to the input conversation text data.

[0122] The first time data is the most recent time when the user inputs dialogue text data through the input module.

[0123] During the entire conversation process, event objects are identified in real time based on the continuously input conversation text. If a new event object is found, the new data is replaced in the event object cache, and the numerical priority of the original event object is lowered. The robot calls the event object's event graph based on the current highest priority (the latest event object in the event object cache) to perform text analysis and generate reply content.

[0124] For further explanation of the present invention, see Figure 1 , 2 , 3, 4, 5, a receiving module, which is used for the user to receive reply content data;

[0125] The step of "calling the cause-and-effect graph corresponding to the event object and generating reply content through text analysis" is: calling the cause-and-effect graph corresponding to the priority event object, combining the logical operation method, outputting the reply content data and sending it to the receiving module.

[0126] The present invention combines the event graph and the logic operation method in the above manner to quickly form reply data, generate the data as a speech and return it to the front-end user for display.

[0127] Among them, the logical operation method can adopt Boolean operation, which is a digital symbolized logical deduction method, including union, intersection and subtraction. This logical operation method is used in graphic processing operations to combine simple basic graphics to produce new shapes, and develop from two-dimensional Boolean operation to three-dimensional graphics Boolean operation.

[0128] Among them, when the robot has the ability to track event objects and reason about reasoning graphs, the entire conversation interaction process can locate the topic according to the customer's conversation text, and through logical reasoning, it can perform reasoning relationship calculations and Boolean calculations on the unresolved issues in the topic to form reply data, and generate the data as a speech and return it to the front-end user for display.

[0129] For further explanation of the present invention, see Figure 1 , 2 , 3, 4, 5, the input module is: mobile phone, computer.

[0130] For further explanation of the present invention, see Figure 1 , 2 , 3, 4, 5, the processor is electrically connected to the first building module and the second building module.

[0131] The embodiments described above are merely descriptions of preferred implementations of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope of the present invention.

Claims

1. A dialogue control system based on subject-object tracking and cognitive reasoning, characterized in that: include The first construction module constructs an event object database through a business rule vocabulary and a product vocabulary; A second construction module, which constructs a matter graph corresponding to the event object by using the first method; An input module, which is used for users to input dialogue text data; A recognition module that recognizes an event object by comparing the conversation text data with the event object; A tracking module, which is used to track event objects in real time; A processor, which constructs an event object database through a first construction module, and constructs a matter graph corresponding to the event object through a second construction module; When the user inputs a conversation text, the processor controls the recognition module to compare the word vector value of the conversation text data with the word vector value of the event object to identify the event object, and the processor controls the tracking module to track the event object in real time, and calls the event graph corresponding to the event object to generate reply content through text analysis; include A first database pre-stored with knowledge base text data; A business rule word library, which is used to store the first key word and the second key word; A second database pre-stored with relational database information; Product word library, which is used to store the name information corresponding to the primary key; The step of "a first construction module, which constructs an event object database through a business rule word library and a product word library" is: the processor uses a new word discovery algorithm of a text processing tool to calculate the information entropy and mutual information of the knowledge base text data in the first database to mine unregistered words, and add them to the first database; In the top-level node of the knowledge base text data in the first database, the processor uses a first extraction method to extract a first key word and adds it to the business rule vocabulary; At the end node of the knowledge base text data in the first database, the processor uses a second extraction method to extract a second key word and adds it to the business rule word library; The processor extracts relational database information from the second database using an information extraction method, extracts name information corresponding to the primary key, and adds the name information to the product word library; The processor controls the first construction module to merge the business rule vocabulary and the product vocabulary into an event object database.

2. A dialogue control system based on subject-object tracking and cognitive reasoning according to claim 1, characterized in that: The first database also pre-stores a knowledge graph; The first method is that the processor can refer to the hierarchical relationship of entity data in the knowledge graph to construct a logic graph of hierarchical reasoning relationship.

3. A dialogue control system based on subject object tracking and cognitive reasoning according to claim 1, characterized in that: The second database also pre-stores unstructured text; The first method is that the processor can identify the relational database information and unstructured text in the second database through a relational learning model, and construct a logical graph of constituent reasoning, reverse reasoning, and concurrent reasoning relationships.

4. A dialogue control system based on subject-object tracking and cognitive reasoning according to claim 2, characterized in that: The steps of the “recognition module, which identifies the event object by comparing the conversation text data and the event object” are: Based on the event object database, the processor calculates the first word vector value of each event object data in the event object database through a vector calculation model; The processor is capable of converting conversation text data into an input event object, and calculating a second word vector value corresponding to the input event object through a vector calculation model. The processor uses a vector similarity calculation method to determine whether the similarity between the first word vector value and the second word vector value is greater than a first preset threshold. If so, the input event object is determined to be the event object with a similarity greater than the first preset threshold, and is stored in an event object database. If not, the event object is cached through a regular rule method.

5. A dialogue control system based on subject-object tracking and cognitive reasoning according to claim 4, characterized in that: The step of "tracking module, which is used to track the event object in real time" is: the processor compares the conversation text data and the event object in real time through the recognition module to identify the event object; The processor can obtain first time data of the dialogue text data input by the user through the input module, and convert the event object corresponding to the dialogue text data corresponding to the first time data into a priority event object.

6. A dialogue control system based on subject-object tracking and cognitive reasoning according to claim 5, characterized in that: Also includes A receiving module, which is used for users to receive reply content data; The step of "calling the cause-and-effect graph corresponding to the event object and generating reply content through text analysis" is: calling the cause-and-effect graph corresponding to the priority event object, combining the logical operation method, outputting the reply content data and sending it to the receiving module.

Images