CN112487810A - Intelligent customer service method, device, equipment and storage medium - Google Patents
Intelligent customer service method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN112487810A CN112487810A CN202011496461.4A CN202011496461A CN112487810A CN 112487810 A CN112487810 A CN 112487810A CN 202011496461 A CN202011496461 A CN 202011496461A CN 112487810 A CN112487810 A CN 112487810A
- Authority
- CN
- China
- Prior art keywords
- question
- user
- matching
- standard
- retrieval
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3343—Query execution using phonetics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
- G06Q30/016—After-sales
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Molecular Biology (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Finance (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an intelligent customer service method, which identifies and divides a semantic field of a question consulted by a user and further determines user intention in the attributive field, so that retrieval of a user standard question can be realized according to a question retrieval algorithm corresponding to the corresponding intention in the corresponding semantic field, flow configuration is carried out according to different business requirements, reliable response of multiple businesses under a complex application scene is guaranteed, a pre-configured knowledge base is called to determine answer corpora corresponding to the standard question, and the answer corpora is fed back to a user side, so that conversation requirements of the user are met, and user experience is improved. The invention also discloses an intelligent customer service device, equipment and a readable storage medium, and has corresponding technical effects.
Description
Technical Field
The invention relates to the technical field of electronics, in particular to an intelligent customer service method, an intelligent customer service device, intelligent customer service equipment and a readable storage medium.
Background
In order to facilitate quick response to user inquiry about problems in fields such as tax field, business service and the like at any time so as to meet daily inquiry requirements of users, a developer-oriented session system supports realization of intelligent sessions based on natural language processing on different message terminals at present.
With the development of services in the application field, the service scenes applied by the intelligent session system are numerous and complex, the service scenes in the vertical field are finely divided, the coupling between the services is serious, and for the consultation of users, the current intelligent session system mostly realizes the response of user requests through end-to-end single-turn conversation, the problem that the unified solution cannot be realized through an end-to-end mode due to the complexity of the service scenes is solved, different service scenes need to be combined by one or more algorithms, the algorithms need to be uniformly scheduled and managed, and the current session response mechanism is difficult to effectively solve the daily consultation problem of the clients, so that the user experience is poor.
In summary, how to accurately determine the user consultation service problem and make an accurate response so as to improve the user experience is a technical problem that needs to be solved by those skilled in the art.
Disclosure of Invention
The invention aims to provide an intelligent customer service method, an intelligent customer service device, intelligent customer service equipment and a readable storage medium, which can accurately determine a user consultation service problem and make an accurate response.
In order to solve the technical problems, the invention provides the following technical scheme:
an intelligent customer service method, comprising:
after receiving a question consulted by a user, calling a domain classification model to identify semantic domain features in the question to obtain a target domain;
calling an intention classification model to identify semantic intention characteristics in the problem to obtain the user intention in the target field;
calling a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval, and determining a standard question corresponding to the question;
and calling a pre-configured knowledge base to determine an answer corpus corresponding to the standard question, and feeding back the answer corpus to the user side.
Optionally, invoking a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval, and determining a standard question corresponding to the question includes:
when the user intention is a common question, calling an ElasticSearch retrieval algorithm to retrieve a standard question corresponding to the question, and taking an obtained retrieval result as a rough recall result;
and calling an edit distance algorithm to evaluate the accuracy of the coarse recall result, and taking the coarse recall result with the highest accuracy as a standard question.
Optionally, invoking a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval, and determining a standard question corresponding to the question includes:
when the user intention is a chatting question, calling an ElasticSearch search algorithm to search a standard question corresponding to the question, and taking an obtained search result as a rough recall result;
calling an edit distance algorithm to perform precision evaluation on the coarse recall result, and acquiring the coarse recall result with the highest precision as a retrieval result;
judging whether the precision corresponding to the retrieval result reaches a preset precision matching threshold value or not;
if so, taking the retrieval result as a standard question;
if not, the problem is subjected to generalized conversion by adopting a TextCNN multi-classification model, and a result output by the TextCNN multi-classification model is used as a standard question.
Optionally, invoking a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval, and determining a standard question corresponding to the question includes:
when the user intention is a business problem, extracting basic features of the problem; wherein the base features include: word segmentation vectors, parts of speech and semantics;
obtaining a dialog flow corresponding to the problem according to the basic characteristics;
judging whether historical session information exists in the cache; the historical session information comprises the completed slot position information and the last session interruption position;
if yes, extracting slot bit information according to the historical conversation information and the conversation flow;
and matching a corresponding standard question according to the slot position information.
Optionally, obtaining a dialog flow corresponding to the question according to the basic feature includes:
matching the question and the word segmentation vector in a template library by using collected condition information;
if the condition information is obtained through matching, the condition information obtained through matching is used as a dialogue flow corresponding to the problem;
if the question is not matched with the dialog flow, calling an ElasticSearch retrieval algorithm to match the dialog flow corresponding to the question;
calling an edit distance algorithm to perform precision evaluation on the matching dialogue stream obtained by matching the ElasticSearch search algorithm, and obtaining the matching dialogue stream with the highest precision;
judging whether the precision corresponding to the matching dialogue flow with the highest precision reaches the precision matching threshold value;
if so, taking the matching conversation flow with the highest precision as a conversation flow corresponding to the problem;
if not, predicting the problem by adopting a TextCNN + MLP multi-classification fusion model to obtain a prediction result and corresponding similarity;
if the similarity is not smaller than a preset threshold value, taking the prediction result as a dialog flow corresponding to the problem;
and if the similarity is smaller than the preset threshold value, starting manual processing.
Optionally, matching a corresponding standard question according to the slot position information includes:
performing combined operation on the slot position information by using a Drools rule base to obtain combined information;
judging whether the combined information has a corresponding standard question or not;
if so, taking the standard question corresponding to the combined information as the standard question obtained by matching;
if not, determining a standard question meeting the slot position information at most;
determining the slot position information which is not collected and corresponds to the standard questions;
generating a user question according to the slot position information which is not collected, and outputting the user question;
updating the slot position information according to the received information replied by the user to obtain updated slot position information;
and performing the step of performing combined operation on the slot position information by using a Drools rule base, wherein the slot position information is the updated slot position information.
Optionally, before matching the corresponding standard according to the slot information, the method further includes:
judging whether an incomplete slot position exists or not;
and if so, acquiring the slot position information corresponding to the unrepleted slot position.
An intelligent customer service device, the device comprising:
the domain determining unit is used for calling a domain classification model to identify semantic domain features in the problem after receiving the problem consulted by the user to obtain a target domain;
the intention determining unit is used for calling an intention classification model to identify semantic intention characteristics in the problem and obtaining the user intention in the target field;
the standard question searching unit is used for calling a question searching algorithm corresponding to the user intention to perform user standard question searching and determining a standard question corresponding to the question;
and the corpus output unit is used for calling a pre-configured knowledge base to determine the answer corpus corresponding to the standard question and feeding back the answer corpus to the user side.
According to the intelligent customer service method provided by the invention, the semantic field is identified and divided for the questions consulted by the user, the intention of the user is further determined in the attributive field, so that the retrieval of the user standard questions can be realized according to the question retrieval algorithm corresponding to the corresponding intention in the corresponding semantic field, the process configuration is carried out aiming at different business requirements, the reliable response of multiple services under the complex application scene can be met, then the pre-configured knowledge base is called to determine the answer corpus corresponding to the standard questions, and the answer corpus is fed back to the user side, so that the conversation requirement of the user is met, and the user experience is improved.
Correspondingly, the invention also provides an intelligent customer service device, equipment and a readable storage medium corresponding to the intelligent customer service method, which have the technical effects and are not described herein again.
Drawings
In order to more clearly illustrate the embodiments of the present invention or technical solutions in related arts, the drawings used in the description of the embodiments or related arts will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flowchart illustrating an embodiment of an intelligent customer service method;
FIG. 2 is a schematic diagram of an overall implementation architecture of an intelligent customer service in the tax domain according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a matching implementation flow of a service question standard question in the embodiment of the present invention;
FIG. 4 is a schematic structural diagram of an intelligent customer service device according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an intelligent customer service device in an embodiment of the present invention.
Detailed Description
The core of the invention is to provide an intelligent customer service method which can accurately determine the problem of the user consultation service and make an accurate response.
In order that those skilled in the art will better understand the disclosure, the invention will be described in further detail with reference to the accompanying drawings and specific embodiments. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1, fig. 1 is a flowchart of an intelligent customer service method according to an embodiment of the present invention, where the method includes the following steps:
s101, after receiving a question consulted by a user, calling a domain classification model to identify semantic domain features in the question to obtain a target domain;
firstly, a user question is divided into one of the service fields according to a service field (the number and the type of the service fields are not limited) which is configured in advance through a semantic feature field classification model. In this embodiment, semantic domain features are divided according to a service application scenario, so as to call a semantic feature analysis package (or component) in a corresponding domain to implement accurate analysis on a user problem.
Taking the tax field as an example, the service field may be divided according to the classification of tax methods, for example, the service field may be divided into 13 tax method division fields (including personal income tax, value-added tax, enterprise income tax, etc.), and then the question consulted by the user is divided into one of the 13 fields, so as to invoke the information configured in the corresponding field (such as personal income tax) to perform accurate question identification in the field.
The semantic domain features are identified through a pre-trained domain classification model, specific model types, model structures and training modes of the domain classification model are not limited in the embodiment, in order to improve identification accuracy, the semantic domain features in the domain classification model identification problem constructed based on the TextCNN + MLP fusion model can be called, after verification, samples in corresponding domains are input into the model to be trained, the domain feature classification effect is better based on the TextCNN + MLP fusion model, in the embodiment, only the model types are used as examples for introduction, other model types can refer to the introduction of the embodiment, and the description is omitted here.
S102, calling an intention classification model to identify semantic intention characteristics in a problem to obtain a user intention in a target field;
the method includes the steps of calling an intention classification model to identify semantic intention features according to semantic intention features asked by users, and judging user intentions (such as chatting questions, service questions and common questions (FAQ)) so that accurate identification can be achieved according to corresponding user question retrieval algorithms.
In addition, in this embodiment, specific model types, model structures, and training modes of the called pre-trained intent classification model for recognizing the intent of the user are not limited, in order to improve recognition accuracy, a TextCNN three classification model may be used for semantic intent feature recognition, and other models may refer to the description of this embodiment and are not described herein again.
S103, calling a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval, and determining a standard question corresponding to the question;
adaptive problem retrieval algorithms are configured for different user intentions in each semantic field in advance to meet accurate identification analysis under different user intentions, specific algorithm configuration of the problem retrieval algorithm for identifying different user intentions in each field is not limited in this embodiment, and configuration can be performed according to requirements of business parties, for example, multiple matching algorithms can be configured in advance, such as a rule base (for quickly matching user questions of a fixed sentence pattern and triggering entries of different flows in a dialog flow through a combination of a plurality of conditions), an edit distance (a text similarity measurement algorithm), and depth model matching (including depth model algorithms such as CNN multi-classification, FastText multi-classification, Bert multi-classification, and the like).
And S104, calling a pre-configured knowledge base to determine an answer corpus corresponding to the standard question, and feeding the answer corpus back to the user side.
After the standard questions are determined, the corresponding preset answer linguistic data (namely the answers corresponding to the user questions) are obtained from the knowledge base according to the standard questions, and a plurality of standard questions and the corresponding answer linguistic data are preset in the knowledge base, so that the basic question and answer requirements of the user can be met.
For deepening understanding, an overall implementation architecture of an intelligent customer service in a tax field is introduced, and as shown in fig. 2, the overall implementation architecture of the intelligent customer service in the tax field is shown, tax questions input by a user are subjected to field classification through a field classification model to obtain a tax field, user intentions (common questions, chatty questions and leisure questions) are further determined through a three-classification model in the tax field, so that corresponding standard questions can be determined according to standard question matching algorithms corresponding to different user intentions, and after the standard questions are determined, corresponding answer corporations are taken out from a knowledge base after dimension filtering (referring to the basis of questioning, different bases correspond to different answers, and the basis can be province, city, county, region, time, degree, identity and the like, and no limitation is made here).
Based on the above introduction, in the intelligent customer service method provided in this embodiment, by identifying and dividing the semantic field of the questions consulted by the user and further determining the user intention in the attributive field, the retrieval of the user standard questions can be realized according to the question retrieval algorithm corresponding to the corresponding intention in the corresponding semantic field, and by performing flow configuration for different service requirements, it is ensured that the multi-service reliable response can be satisfied in a complex application scene, and then a preconfigured knowledge base is called to determine the answer corpus corresponding to the standard questions, and the answer corpus is fed back to the user side, so that the session requirement of the user is satisfied, and the user experience is improved.
In the above embodiment, a specific implementation manner of invoking the question retrieval algorithm corresponding to the user intention to perform the user standard question retrieval to determine the standard question corresponding to the question is not limited, where the matching algorithm and the corresponding implementation flow may be correspondingly set according to the matching requirement of the actual standard question, and in order to deepen understanding, a retrieval implementation manner of the standard question is introduced here, which is specifically as follows:
1. when the user intention is a common problem, in order to improve the identification accuracy and the identification efficiency aiming at the common problem, a problem retrieval algorithm corresponding to the user intention is called to carry out user standard question retrieval, and the implementation mode of determining the standard question corresponding to the problem is as follows:
(1) when the user intends to be a common problem, calling an ElasticSearch retrieval algorithm to retrieve a standard question corresponding to the problem, and taking an obtained retrieval result as a rough recall result;
the accuracy rate of the ElasticSearch search is not high, but the number of all the matching standard questions obtained by the search is large, the range is wide, and further accurate standard question analysis can be realized according to a large number of search results obtained by the ElasticSearch search.
(2) And calling an edit distance algorithm to perform precision evaluation on the coarse recall result, and taking the coarse recall result with the highest precision as a standard question.
The method comprises the steps of further filtering results of the rough recalls through an edit distance algorithm, wherein the edit distance algorithm can evaluate the distance between questions consulted by a user aiming at each result of the rough recall, namely precision evaluation, can perform threshold value screening according to the evaluation results and rank the results according to the edit distance (for example, the rough recall results exceeding the threshold value can be screened first, then the rough recall results exceeding the threshold value are ranked in the edit distance (from large to small or from small to large), the minimum value is determined, or the edit distances of all the rough recall results can be directly ranked, the minimum value is determined), and the rough recall result corresponding to the minimum distance (namely the rough recall result with the highest precision) is used as a standard question.
Because the question-asking modes corresponding to the common questions are many and the retrieval difficulty of the standard questions is high, the standard question retrieval mode for the common questions provided in the embodiment firstly searches results in a large range through the ElasticSearch retrieval, and then locates the accurate standard questions according to the edit distance algorithm based on the rough recall result, so that effective standard question matching retrieval can be realized.
2. When the user intention is a chat question, in order to improve the recognition accuracy and recognition efficiency for the chat question, a question retrieval algorithm corresponding to the user intention is called to perform user standard question retrieval, and the implementation mode of determining the standard question corresponding to the question is as follows:
(1) when the user intends to be a chatting question, calling an ElasticSearch search algorithm to search a standard question corresponding to the question, and taking an obtained search result as a rough recall result;
the accuracy rate of the ElasticSearch search is not high, but the number of all the matching standard questions obtained by the search is large, the range is wide, and further accurate standard question analysis can be realized according to a large number of search results obtained by the ElasticSearch search.
(2) Calling an edit distance algorithm to perform precision evaluation on the coarse recall result, and acquiring the coarse recall result with the highest precision as a retrieval result;
(3) judging whether the precision corresponding to the retrieval result reaches a preset precision matching threshold value or not;
(4) if so, taking the retrieval result as a standard question;
(5) if not, the problem is subjected to generalization conversion by adopting the TextCNN multi-classification model, and the result output by the TextCNN multi-classification model is used as a standard question.
The question asking forms and the question asking objects of the chatting questions are various and have multiple grammar forms, the TF-IDF algorithm is often adopted to realize the retrieval of the standard questions in the traditional mode, but the retrieval accuracy is lower aiming at the various language forms of the chatting questions. In order to solve the above problems and improve the retrieval accuracy of the standard questions, in this embodiment, it is proposed to perform coarse filtering by using an elastic search retrieval, perform approximate exact matching according to an edit distance algorithm (an exact matching threshold may be set to a matching value corresponding to an error of one word, for example), and if the approximate exact matching is not matched, perform generalized transformation by using a TextCNN multi-classification model problem, and further determine the sub-intentions of the user. The method can ensure accurate retrieval in the multi-question grammar form.
3. When the user intention is a business problem, in order to improve the identification accuracy and the identification efficiency under the business problem, a problem retrieval algorithm corresponding to the user intention is called to carry out user standard question retrieval, and the implementation mode of determining the standard question corresponding to the problem is as follows:
(1) when the user intention is a business problem, extracting basic features of the problem; wherein, the basic features include: word segmentation vectors, parts of speech and semantics;
the basic feature extraction mainly refers to extracting word segmentation vectors, part of speech tagging and semantic analysis of a problem, and may further extract other feature information, which is not limited in this embodiment. In addition, in this embodiment, an implementation manner of the basic feature extraction is not limited, and an implementation manner of a related technology may be referred to, for example, segmentation, part-of-speech tagging and semantic analysis may be performed on the user question by using the final segmentation, and a word vector of each segmentation in the user question is obtained by combining word vectors of encyclopedic. In this embodiment, only the above implementation is taken as an example, and other implementations are not described herein again.
(2) Obtaining a dialog flow corresponding to the problem according to the basic characteristics;
the method is divided according to service scenes, one conversation flow comprises a plurality of scenes, one scene belongs to one conversation flow intelligently, the conversation flow is conversation data, and the conversation flow can indicate user intention. The specific concept of conversation flow can refer to the introduction in the related intelligent customer service system, and is not described in detail herein. In this embodiment, the matching of the conversation flow is obtained through the basic features displayed in the problem, and the matching accuracy can be improved compared with the method that the matching is directly obtained based on the problem data, and the matching accuracy is improved. The implementation process of performing matching acquisition on the dialog flow based on the basic features is not limited in this embodiment, and one or more matching algorithms may be selected for matching, for example, one or more of a rule base, an elastic search, an edit distance algorithm, and multi-classification fusion prediction may be selected to implement matching of the dialog flow.
The implementation steps of obtaining the dialog flow corresponding to the question according to the basic characteristics are as follows:
(2.1) matching the question and the word segmentation vector in a template library by using the collected condition information;
matching the user question and word segmentation information in a configured template library, then matching the collected condition information by using a rule library (Drools), and if the condition information is not matched, executing the next dialog flow matching step; if the matching is achieved, the condition information obtained by matching can be directly used as the dialogue flow corresponding to the question. The process of matching according to the collected condition information based on the template library is weak in generalization capability, but the realization efficiency is high, and the matching step can be executed firstly to improve the realization efficiency.
(2.2) if the condition information is obtained through matching, taking the condition information obtained through matching as a dialogue flow corresponding to the problem;
(2.3) if the matching is not achieved, calling an ElasticSearch retrieval algorithm to match the dialogue flow corresponding to the problem;
the accuracy rate of the ElasticSearch search is not high, but the number of all the matching standard questions obtained by the search is large, the range is wide, and further accurate standard question analysis can be realized according to a large number of search results obtained by the ElasticSearch search.
(2.4) calling an edit distance algorithm to perform precision evaluation on the matched dialogue flow obtained by matching the ElasticSearch search algorithm, and obtaining the matched dialogue flow with the highest precision;
(2.5) judging whether the precision corresponding to the matching conversation flow with the highest precision reaches a precision matching threshold value;
(2.6) if the matching conversation flow with the highest precision is achieved, taking the matching conversation flow with the highest precision as a conversation flow corresponding to the problem;
the precision of approximate accurate matching by adopting the ElasticSearch search and edit distance algorithm is higher than that of a rule matching mode, and more accurate dialogue flow matching can be realized.
(2.7) if the problem does not reach the preset value, predicting the problem by adopting a TextCNN + MLP multi-classification fusion model to obtain a prediction result and corresponding similarity;
(2.8) if the similarity is not less than a preset threshold, taking the prediction result as a dialog flow corresponding to the problem;
and (2.9) if the similarity is smaller than a preset threshold value, starting manual processing.
And predicting the user questions by adopting a TextCNN + MLP multi-classification fusion model, and directly performing manual processing if the similarity obtained by the model is smaller than a configured threshold value.
In this embodiment, the matching process configured by the three matching modes is taken as an example for introduction, and the accuracy is higher and higher under the configuration, so that the accurate matching can be guaranteed while the implementation efficiency is guaranteed. For other implementation manners, reference may be made to the description of the embodiment, and details are not described herein.
(3) Judging whether historical session information exists in the cache;
according to the historical session information before the current session information is recovered from the cache, the historical session information may include previously completed slot position (after the user intention is preliminarily clarified, necessary information is acquired to finally obtain an explicit user instruction), last session interruption position, and the like.
(4) If so, extracting slot position information according to the historical session information and the conversation flow;
the slot position information is extracted, and the slot position information extraction implementation manner is not limited in this embodiment and can be implemented based on a conventional slot position information extraction algorithm. One way of implementation is as follows: if multiple rounds of conversations are in a cold start stage (the language material and the data volume are less), slot position information can be directly extracted through text multi-classification models such as FastText and TextCNN; when the slot position is not in a cold start stage, namely the data volume is accumulated to a certain degree, a Bert + CRF model can be adopted to identify NER, and the accuracy of slot position extraction is further improved. In this embodiment, only the above implementation manner is described as an example, and other extraction implementation manners can refer to the description of this embodiment, which is not described herein again.
If no historical session information exists, slot information can be directly extracted according to the current user question, and the implementation process can refer to the above description, which is not limited herein.
(5) And matching the corresponding standard questions according to the slot position information.
The slot positions correspond to the slot position information, the corresponding standard questions under the slot position information are matched according to the information of the slot positions matched with the session information of the user, the standard questions are pre-configured questions with standard answers, and the received user questions can be searched by matching the corresponding standard questions. However, in this embodiment, the implementation manner of matching the corresponding standard questions according to the slot position information is not limited, and for the sake of deeper understanding, an implementation manner of matching the corresponding standard questions according to the slot position information is introduced here, as follows:
(5.1) performing combined operation on the slot position information by using a Drools rule base to obtain combined information;
and combining the operation information corresponding to all the current slot positions, performing combined operation on the information corresponding to each slot position by using a Drools rule base, and judging whether the combined value corresponds to a standard problem.
(5.2) judging whether the combined information has a corresponding standard question or not;
(5.3) if the combined information exists, taking the standard question corresponding to the combined information as the standard question obtained by matching;
(5.4) if the slot position information does not exist, determining that the standard of the most slot position information is met;
if the standard problem cannot be obtained according to the current slot position combination, a standard question which can be met only by filling the slot position at least is needed to be found, slot position information which is not collected yet is found (according to slot position characteristics, priority, filling property and the like), and a user is asked reversely;
(5.5) determining the slot position information which is not collected and corresponds to the standard;
(5.6) generating a user question according to the slot position information which is not collected, and outputting the user question;
(5.7) updating the slot position information according to the received information replied by the user to obtain updated slot position information;
and updating the slot position information and the session information in the cache.
And (5.8) performing a combined operation on the slot position information by using the Drools rule base, wherein the slot position information is updated slot position information.
In order to increase the slot position information, so that more accurate matching of standard questions is realized based on richer slot position information, before matching corresponding standard questions according to the slot position information, whether uncompensated slot positions exist can be further judged according to currently configured conversation flow information, whether certain slot positions need to be complemented is determined, if uncompensated operation exists, slot position information corresponding to the uncompensated slot positions is obtained, specifically, the obtaining mode of the slot position information is not limited, the slot position information corresponding to the uncompensated slot positions can be extracted from an associated system, and the slot position information can be complemented according to answers of users in a mode of asking back the users, and is not repeated here.
As shown in fig. 3, a schematic diagram of a matching implementation process of a standard question of a business question is provided, in which, for a business question, corresponding basic features are extracted first, then, a user intention (sub-intention) is identified according to the extracted basic features (according to rule base matching, edit distance matching, and semantic model matching), extraction and state updating of a dialog flow are performed based on the identified user intention, and a corresponding answer corpus is determined according to the matched dialog flow.
Corresponding to the above method embodiment, the embodiment of the present invention further provides an intelligent customer service device, and the below described intelligent customer service device and the above described intelligent customer service method may be referred to correspondingly.
Referring to fig. 4, the apparatus includes the following modules:
the domain determining unit 110 is mainly configured to, after receiving a question consulted by a user, call a domain classification model to identify semantic domain features in the question to obtain a target domain;
the intention determining unit 120 is mainly configured to invoke an intention classification model to identify semantic intention features in a question, so as to obtain a user intention in a target field;
the standard question retrieval unit 130 is mainly used for invoking a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval and determining a standard question corresponding to a question;
the corpus output unit 140 is mainly used to call a pre-configured knowledge base to determine an answer corpus corresponding to the standard question, and feed back the answer corpus to the user side.
Corresponding to the above method embodiment, the embodiment of the present invention further provides an intelligent customer service device, and the following intelligent customer service device and the above described intelligent customer service method may be referred to in correspondence.
This intelligence customer service equipment mainly includes:
a memory for storing a computer program;
and the processor is used for realizing the steps of the intelligent customer service method of the embodiment of the method when executing the computer program.
Specifically, referring to fig. 5, a schematic diagram of a specific structure of an intelligent customer service device provided in this embodiment is provided, where the intelligent customer service device may generate a relatively large difference due to different configurations or performances, and may include one or more processors (CPUs) 322 (e.g., one or more processors) and a memory 332, where the memory 332 stores one or more computer applications 342 or data 344. Memory 332 may be, among other things, transient or persistent storage. The program stored in memory 332 may include one or more modules (not shown), each of which may include a sequence of instructions operating on a data processing device. Still further, the central processor 322 may be configured to communicate with the memory 332 to execute a series of instructional operations on the intelligent customer service device 301 within the memory 332.
The intelligent customer service device 301 may also include one or more power sources 326, one or more wired or wireless network interfaces 350, one or more input-output interfaces 358, and/or one or more operating systems 341.
The steps in the intelligent customer service method described above may be implemented by the structure of an intelligent customer service device.
Corresponding to the above method embodiment, an embodiment of the present invention further provides a readable storage medium, and a readable storage medium described below and an intelligent customer service method described above may be referred to in correspondence.
A readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the intelligent customer service method of the above-mentioned method embodiments.
The readable storage medium may be a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and various other readable storage media capable of storing program codes.
Those of skill would further appreciate that the various illustrative components and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
Claims (10)
1. An intelligent customer service method, characterized by comprising:
after receiving a question consulted by a user, calling a domain classification model to identify semantic domain features in the question to obtain a target domain;
calling an intention classification model to identify semantic intention characteristics in the problem to obtain the user intention in the target field;
calling a question retrieval algorithm corresponding to the user intention to perform user standard question retrieval, and determining a standard question corresponding to the question;
and calling a pre-configured knowledge base to determine an answer corpus corresponding to the standard question, and feeding back the answer corpus to the user side.
2. The intelligent customer service method according to claim 1, wherein invoking a question retrieval algorithm corresponding to the user intention to perform a user standard question retrieval and determining a standard question corresponding to the question comprises:
when the user intention is a common question, calling an ElasticSearch retrieval algorithm to retrieve a standard question corresponding to the question, and taking an obtained retrieval result as a rough recall result;
and calling an edit distance algorithm to evaluate the accuracy of the coarse recall result, and taking the coarse recall result with the highest accuracy as a standard question.
3. The intelligent customer service method according to claim 1, wherein invoking a question retrieval algorithm corresponding to the user intention to perform a user standard question retrieval and determining a standard question corresponding to the question comprises:
when the user intention is a chatting question, calling an ElasticSearch search algorithm to search a standard question corresponding to the question, and taking an obtained search result as a rough recall result;
calling an edit distance algorithm to perform precision evaluation on the coarse recall result, and acquiring the coarse recall result with the highest precision as a retrieval result;
judging whether the precision corresponding to the retrieval result reaches a preset precision matching threshold value or not;
if so, taking the retrieval result as a standard question;
if not, the problem is subjected to generalized conversion by adopting a TextCNN multi-classification model, and a result output by the TextCNN multi-classification model is used as a standard question.
4. The intelligent customer service method according to claim 1, wherein invoking a question retrieval algorithm corresponding to the user intention to perform a user standard question retrieval and determining a standard question corresponding to the question comprises:
when the user intention is a business problem, extracting basic features of the problem; wherein the base features include: word segmentation vectors, parts of speech and semantics;
obtaining a dialog flow corresponding to the problem according to the basic characteristics;
judging whether historical session information exists in the cache; the historical session information comprises the completed slot position information and the last session interruption position;
if yes, extracting slot bit information according to the historical conversation information and the conversation flow;
and matching a corresponding standard question according to the slot position information.
5. The intelligent customer service method according to claim 4, wherein obtaining the dialog flow corresponding to the question according to the base feature comprises:
matching the question and the word segmentation vector in a template library by using collected condition information;
if the condition information is obtained through matching, the condition information obtained through matching is used as a dialogue flow corresponding to the problem;
if the question is not matched with the dialog flow, calling an ElasticSearch retrieval algorithm to match the dialog flow corresponding to the question;
calling an edit distance algorithm to perform precision evaluation on the matching dialogue stream obtained by matching the ElasticSearch search algorithm, and obtaining the matching dialogue stream with the highest precision;
judging whether the precision corresponding to the matching dialogue flow with the highest precision reaches the precision matching threshold value;
if so, taking the matching conversation flow with the highest precision as a conversation flow corresponding to the problem;
if not, predicting the problem by adopting a TextCNN + MLP multi-classification fusion model to obtain a prediction result and corresponding similarity;
if the similarity is not smaller than a preset threshold value, taking the prediction result as a dialog flow corresponding to the problem;
and if the similarity is smaller than the preset threshold value, starting manual processing.
6. The intelligent customer service method of claim 4, wherein matching the corresponding criteria based on the slot information comprises:
performing combined operation on the slot position information by using a Drools rule base to obtain combined information;
judging whether the combined information has a corresponding standard question or not;
if so, taking the standard question corresponding to the combined information as the standard question obtained by matching;
if not, determining a standard question meeting the slot position information at most;
determining the slot position information which is not collected and corresponds to the standard questions;
generating a user question according to the slot position information which is not collected, and outputting the user question;
updating the slot position information according to the received information replied by the user to obtain updated slot position information;
and performing the step of performing combined operation on the slot position information by using a Drools rule base, wherein the slot position information is the updated slot position information.
7. The intelligent customer service method of claim 4, wherein before matching the corresponding criteria according to the slot information, further comprising:
judging whether an incomplete slot position exists or not;
and if so, acquiring the slot position information corresponding to the unrepleted slot position.
8. An intelligent customer service device, the device comprising:
the domain determining unit is used for calling a domain classification model to identify semantic domain features in the problem after receiving the problem consulted by the user to obtain a target domain;
the intention determining unit is used for calling an intention classification model to identify semantic intention characteristics in the problem and obtaining the user intention in the target field;
the standard question searching unit is used for calling a question searching algorithm corresponding to the user intention to perform user standard question searching and determining a standard question corresponding to the question;
and the corpus output unit is used for calling a pre-configured knowledge base to determine the answer corpus corresponding to the standard question and feeding back the answer corpus to the user side.
9. An intelligent customer service device, comprising:
a memory for storing a computer program;
processor for implementing the steps of the intelligent customer service method according to any one of claims 1 to 7 when executing said computer program.
10. A readable storage medium, characterized in that the readable storage medium has stored thereon a computer program which, when being executed by a processor, carries out the steps of the intelligent customer service method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011496461.4A CN112487810B (en) | 2020-12-17 | 2020-12-17 | Intelligent customer service method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011496461.4A CN112487810B (en) | 2020-12-17 | 2020-12-17 | Intelligent customer service method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112487810A true CN112487810A (en) | 2021-03-12 |
| CN112487810B CN112487810B (en) | 2022-03-11 |
Family
ID=74917324
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011496461.4A Active CN112487810B (en) | 2020-12-17 | 2020-12-17 | Intelligent customer service method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112487810B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139058A (en) * | 2021-05-11 | 2021-07-20 | 支付宝(杭州)信息技术有限公司 | User obstacle identification method and system |
| CN113434656A (en) * | 2021-07-21 | 2021-09-24 | 广州华多网络科技有限公司 | E-commerce customer service matching method and corresponding device, equipment and medium thereof |
| CN113656572A (en) * | 2021-08-26 | 2021-11-16 | 支付宝(杭州)信息技术有限公司 | Conversation processing method and system |
| CN113821621A (en) * | 2021-09-28 | 2021-12-21 | 中电万维信息技术有限责任公司 | Open intelligent customer service system based on deep learning |
| CN114004581A (en) * | 2021-10-12 | 2022-02-01 | 正数网络技术有限公司 | Intention interaction system based on multi-dimensional government affair knowledge base |
| CN114756663A (en) * | 2022-03-29 | 2022-07-15 | 税友信息技术有限公司 | Intelligent question answering method, system, equipment and computer readable storage medium |
| CN114970559A (en) * | 2022-05-18 | 2022-08-30 | 马上消费金融股份有限公司 | Intelligent response method and device |
| CN115098656A (en) * | 2022-06-27 | 2022-09-23 | 招商银行股份有限公司 | Customer service engine management method, device, system, terminal equipment and storage medium |
| CN118245589A (en) * | 2024-05-28 | 2024-06-25 | 浙江口碑网络技术有限公司 | Task type dialogue method, storage medium and computer device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060301A1 (en) * | 2016-08-31 | 2018-03-01 | Microsoft Technology Licensing, Llc | End-to-end learning of dialogue agents for information access |
| CN110059193A (en) * | 2019-06-21 | 2019-07-26 | 南京擎盾信息科技有限公司 | Legal advice system based on law semanteme part and document big data statistical analysis |
| CN110188204A (en) * | 2019-06-11 | 2019-08-30 | 腾讯科技（深圳）有限公司 | A kind of extension corpora mining method, apparatus, server and storage medium |
| CN111831795A (en) * | 2019-04-11 | 2020-10-27 | 北京猎户星空科技有限公司 | Multi-turn conversation processing method and device, electronic equipment and storage medium |
| CN112069298A (en) * | 2020-07-31 | 2020-12-11 | 杭州远传新业科技有限公司 | Human-computer interaction method, device and medium based on semantic web and intention recognition |
-
2020
- 2020-12-17 CN CN202011496461.4A patent/CN112487810B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180060301A1 (en) * | 2016-08-31 | 2018-03-01 | Microsoft Technology Licensing, Llc | End-to-end learning of dialogue agents for information access |
| CN111831795A (en) * | 2019-04-11 | 2020-10-27 | 北京猎户星空科技有限公司 | Multi-turn conversation processing method and device, electronic equipment and storage medium |
| CN110188204A (en) * | 2019-06-11 | 2019-08-30 | 腾讯科技（深圳）有限公司 | A kind of extension corpora mining method, apparatus, server and storage medium |
| CN110059193A (en) * | 2019-06-21 | 2019-07-26 | 南京擎盾信息科技有限公司 | Legal advice system based on law semanteme part and document big data statistical analysis |
| CN112069298A (en) * | 2020-07-31 | 2020-12-11 | 杭州远传新业科技有限公司 | Human-computer interaction method, device and medium based on semantic web and intention recognition |
Non-Patent Citations (2)
| Title |
|---|
| JH CURTIS: "什么是NLP -- NLP在人工智能客服中的应用", 《HTTPS://BLOG.CSDN.NET/WEIXIN_43484983/ARTICLE/DETAILS/104537812》 * |
| 王东升等: "面向限定领域问答系统的自然语言理解方法综述", 《计算机科学》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139058A (en) * | 2021-05-11 | 2021-07-20 | 支付宝(杭州)信息技术有限公司 | User obstacle identification method and system |
| CN113434656B (en) * | 2021-07-21 | 2023-04-25 | 广州华多网络科技有限公司 | E-commerce customer service matching method and corresponding device, equipment and medium thereof |
| CN113434656A (en) * | 2021-07-21 | 2021-09-24 | 广州华多网络科技有限公司 | E-commerce customer service matching method and corresponding device, equipment and medium thereof |
| CN113656572A (en) * | 2021-08-26 | 2021-11-16 | 支付宝(杭州)信息技术有限公司 | Conversation processing method and system |
| CN113821621B (en) * | 2021-09-28 | 2023-10-27 | 中电万维信息技术有限责任公司 | Open intelligent customer service system based on deep learning |
| CN113821621A (en) * | 2021-09-28 | 2021-12-21 | 中电万维信息技术有限责任公司 | Open intelligent customer service system based on deep learning |
| CN114004581A (en) * | 2021-10-12 | 2022-02-01 | 正数网络技术有限公司 | Intention interaction system based on multi-dimensional government affair knowledge base |
| CN114756663A (en) * | 2022-03-29 | 2022-07-15 | 税友信息技术有限公司 | Intelligent question answering method, system, equipment and computer readable storage medium |
| CN114970559A (en) * | 2022-05-18 | 2022-08-30 | 马上消费金融股份有限公司 | Intelligent response method and device |
| CN114970559B (en) * | 2022-05-18 | 2024-02-02 | 马上消费金融股份有限公司 | Intelligent response method and device |
| CN115098656A (en) * | 2022-06-27 | 2022-09-23 | 招商银行股份有限公司 | Customer service engine management method, device, system, terminal equipment and storage medium |
| CN118245589A (en) * | 2024-05-28 | 2024-06-25 | 浙江口碑网络技术有限公司 | Task type dialogue method, storage medium and computer device |
| CN118245589B (en) * | 2024-05-28 | 2024-08-27 | 浙江口碑网络技术有限公司 | Task type dialogue method, storage medium and computer device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112487810B (en) | 2022-03-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112487810B (en) | Intelligent customer service method, device, equipment and storage medium | |
| CN109918673B (en) | Semantic arbitration method and device, electronic equipment and computer-readable storage medium | |
| CN110111780B (en) | Data processing method and server | |
| CN110263144A (en) | A kind of answer acquisition methods and device | |
| CN110364146B (en) | Speech recognition method, speech recognition device, speech recognition apparatus, and storage medium | |
| CN112035599B (en) | Query method and device based on vertical search, computer equipment and storage medium | |
| CN109325040B (en) | FAQ question-answer library generalization method, device and equipment | |
| CN112948534A (en) | Interaction method and system for intelligent man-machine conversation and electronic equipment | |
| CN111368042A (en) | Intelligent question and answer method and device, computer equipment and computer storage medium | |
| CN112818109B (en) | Intelligent reply method, medium, device and computing equipment for mail | |
| CN112784590B (en) | Text processing method and device | |
| CN117609444B (en) | Searching question-answering method based on large model | |
| CN117573834B (en) | Multi-robot dialogue method and system for software-oriented instant service platform | |
| CN111159334A (en) | Method and system for house source follow-up information processing | |
| CN111353026A (en) | Intelligent law attorney assistant customer service system | |
| CN112364622A (en) | Dialog text analysis method, dialog text analysis device, electronic device and storage medium | |
| CN116150306A (en) | Training method of question-answering robot, question-answering method and device | |
| CN110795547A (en) | Text recognition method and related product | |
| CN114328913A (en) | Text classification method and device, computer equipment and storage medium | |
| CN117312521A (en) | Processing method for intelligent customer service dialogue and related products | |
| CN117332062A (en) | Data processing method and related device | |
| CN114491010A (en) | Training method and device of information extraction model | |
| CN116186259A (en) | Session cue scoring method, device, equipment and storage medium | |
| CN113570324A (en) | Outbound flow editing method and device, electronic equipment and storage medium | |
| CN112395402A (en) | Depth model-based recommended word generation method and device and computer equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |