CN111143533A - Customer service method and system based on user behavior data - Google Patents
Customer service method and system based on user behavior data Download PDFInfo
- Publication number
- CN111143533A CN111143533A CN201911365582.2A CN201911365582A CN111143533A CN 111143533 A CN111143533 A CN 111143533A CN 201911365582 A CN201911365582 A CN 201911365582A CN 111143533 A CN111143533 A CN 111143533A
- Authority
- CN
- China
- Prior art keywords
- knowledge point
- tree
- knowledge
- category
- point category
- 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
- 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
- G06F16/31—Indexing; Data structures therefor; Storage structures
- G06F16/316—Indexing structures
- G06F16/322—Trees
-
- 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
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Human Computer Interaction (AREA)
- Software Systems (AREA)
- Artificial Intelligence (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 embodiment of the invention discloses a customer service method and a customer service system based on user behavior data, wherein a knowledge point class weight tree is obtained according to the user behavior data and a knowledge base system; acquiring word vector characteristic data according to a problem input by a user; acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data; and according to the knowledge point category, carrying out similar question matching on the knowledge points belonging to the knowledge point category to obtain the question answers input by the user. The method is based on a weight division correction method of user behavior data, information is supplemented for knowledge point identification in a customer service processing process, and accuracy of knowledge point classification is improved.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to a customer service method and a customer service system based on user behavior data.
Background
With the development of natural language processing and computing, intelligent customer service robots play an important role in various industries. At present, most of intelligent customer service systems provide knowledge question and answer service for the vertical business field based on natural language processing, automatic question and answer technology and knowledge base management systems on the basis of a vertical field knowledge base, so that the artificial customer service burden is reduced, the enterprise cost is saved, and the enterprise service efficiency is improved.
The realization of the intelligent customer service system relates to the technologies of text preprocessing, text feature extraction, intention identification, knowledge base retrieval, similarity calculation, intelligent sequencing and the like. The intention recognition judges the service category of the knowledge base which the user wants to consult according to the input problems of the user, then carries out the problem retrieval of the knowledge base in the category, and then obtains the answer which is most matched with the user problems according to the similarity calculation and the intelligent sequencing.
In this process, the accuracy of the intent recognition determines the accuracy of the resulting answer. The intention recognition in a general intelligent customer service robot is a classification model trained on well-classified business knowledge base data, and the classification of a user problem is extracted and predicted according to text features. In the practical application process, due to the diversity of the forms of user questions, the ambiguity of Chinese language and the fact that some businesses often have similar problem dimensions, the intention recognition may not make better predictions in these business categories, for example, in the intelligent customer service of a certain financial institution, when a user asks that "face recognition fails", the intention recognition may involve the problem of account login authentication and the problem in the credit approval process may also involve, because the incomplete expression of the user brings certain difficulty to the intention recognition, and the answer which is possibly replied to the user only through the content matching of the knowledge base is not required by the user.
Disclosure of Invention
The embodiment of the invention provides a customer service method and system based on user behavior data.
In order to achieve the above purpose, the embodiment of the invention adopts the following technical scheme:
in a first aspect, an embodiment of the present invention provides a customer service method based on user behavior data, where a knowledge point category weight tree is obtained according to the user behavior data and a knowledge base system; acquiring word vector characteristic data according to a problem input by a user; acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data; and according to the knowledge point category, carrying out similar question matching on the knowledge points belonging to the knowledge point category to obtain the question answers input by the user.
With reference to the first aspect, in a first possible implementation manner of the first aspect, a user behavior model is established according to user service data collected from a service system; and acquiring a knowledge point category weight tree corresponding to the structure of the knowledge base system according to the knowledge base system and the user behavior model.
With reference to the first possible implementation manner of the first aspect, in a second possible implementation manner of the first aspect, a product service knowledge point category mapping tree is constructed according to a product service system and a knowledge base system of the business system; according to the user behavior model, carrying out quantization processing on the product service knowledge point category mapping tree to obtain a quantized product service knowledge point category mapping tree; constructing a knowledge point category attention tree according to the product service touch; acquiring attention freshness according to attention time of a user to product service; and acquiring a knowledge point category weight tree according to the quantized product service knowledge point category mapping tree, the knowledge point category attention tree and the attention freshness.
With reference to the first possible implementation manner of the first aspect, in a third possible implementation manner of the first aspect, a mapping relationship is established between a product service system of the business system and the knowledge base system, so as to obtain a product service system tree; the product service system tree comprises different product service classifications, and each product service classification comprises different product service subclasses; constructing a knowledge point system tree according to the knowledge base system; the knowledge point system tree comprises different knowledge point classifications, and each knowledge point classification comprises different knowledge point subclasses; constructing a product service knowledge point category mapping tree according to the product service system tree and the knowledge point system tree; the product service knowledge point category mapping tree comprises the knowledge point categories, and each knowledge point category comprises the different product service subclasses.
With reference to the first aspect, in a fourth possible implementation manner of the first aspect, the knowledge point model is trained according to a corpus of a pre-trained knowledge base, so as to obtain a knowledge point identification model; judging the knowledge point category to which the knowledge point belongs according to the word vector characteristic data and the knowledge point identification model, and acquiring knowledge point category information; the knowledge point category information comprises various knowledge point categories and corresponding scores; calculating the difference between the highest-scoring knowledge point category and the next highest-scoring knowledge point category; if the difference is lower than a category confusion threshold, multiplying the weight of the knowledge point category with the highest score corresponding to the knowledge point category weight tree by the highest score to obtain a first value; multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem; and if the difference is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question.
With reference to the first aspect, in a fifth possible implementation manner of the first aspect, cosine distance similarity calculation is performed on knowledge points belonging to the knowledge point category according to TF-IDF features, so as to obtain the first N knowledge points with the highest similarity; and performing semantic similarity calculation on the questions input by the user and the questions of the N knowledge points according to the word vector characteristics, and taking the answer corresponding to the knowledge point with the highest similarity as the answer of the questions input by the user.
In a second aspect, an embodiment of the present invention provides a customer service system based on user behavior data, including:
the knowledge point category weight tree acquisition module is used for acquiring a knowledge point category weight tree according to the user behavior data and the knowledge base system;
the word vector characteristic acquisition module is used for acquiring word vector characteristic data according to the problems input by the user;
the knowledge point category identification module is used for acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data;
and the similar question matching module is used for matching similar questions with knowledge points belonging to the knowledge point category according to the knowledge point category to obtain the question answers input by the user.
With reference to the second aspect, in a first possible implementation manner of the second aspect, the method includes:
the user behavior modeling submodule is used for establishing a user behavior model according to user service data collected from the service system;
and the weight tree obtaining submodule is used for obtaining the knowledge point category weight tree corresponding to the structure of the knowledge base system according to the knowledge base system and the user behavior model.
With reference to the first possible implementation manner of the second aspect, in a second possible implementation manner of the second aspect, the method includes:
the product service knowledge point category mapping tree acquisition submodule is used for constructing a product service knowledge point category mapping tree according to a product service system and a knowledge base system of the service system;
the product service knowledge point category mapping tree quantization submodule is used for carrying out quantization processing on the product service knowledge point category mapping tree according to the user behavior model to obtain a quantized product service knowledge point category mapping tree;
the attention tree obtaining submodule is used for constructing a knowledge point category attention tree according to the product service touch;
the concerned freshness obtaining submodule is used for obtaining concerned freshness according to the concerned time of the user on the product service;
and the weight tree construction submodule is used for acquiring a knowledge point category weight tree according to the quantized product service knowledge point category mapping tree, the knowledge point category attention tree and the attention freshness.
With reference to the first possible implementation manner of the second aspect, in a third possible implementation manner of the second aspect, the method includes:
the product service system tree construction submodule is used for establishing a mapping relation between a product service system of the business system and the knowledge base system to obtain a product service system tree; the product service system tree comprises different product service classifications, and each product service classification comprises different product service subclasses;
the knowledge point system book construction submodule is used for constructing a knowledge point system tree according to the knowledge base system; the knowledge point system tree comprises different knowledge point classifications, and each knowledge point classification comprises different knowledge point subclasses;
the mapping tree construction submodule is used for constructing a product service knowledge point category mapping tree according to the product service system tree and the knowledge point system tree; the product service knowledge point category mapping tree comprises the knowledge point categories, and each knowledge point category comprises the different product service subclasses.
With reference to the second aspect, in a fourth possible implementation manner of the second aspect, the method includes:
the knowledge point model construction submodule is used for carrying out knowledge point model training according to the corpus of a pre-trained knowledge base to obtain a knowledge point identification model;
the knowledge point category information acquisition submodule is used for judging the category of the knowledge point to which the knowledge point belongs according to the word vector characteristic data and the knowledge point identification model and acquiring the category information of the knowledge point; the knowledge point category information comprises various knowledge point categories and corresponding scores;
the knowledge point category identification submodule is used for calculating the difference between the knowledge point category with the highest score and the knowledge point category with the next highest score; if the difference is lower than a category confusion threshold, multiplying the weight of the knowledge point category with the highest score corresponding to the knowledge point category weight tree by the highest score to obtain a first value; multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem; and if the difference is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question.
With reference to the second aspect, in a fifth possible implementation manner of the second aspect, the method includes:
the cosine distance similarity calculation operator module is used for calculating the cosine distance similarity of the knowledge points belonging to the knowledge point category according to the TF-IDF characteristics to obtain the first N knowledge points with the highest similarity;
and the semantic similarity calculation operator module is used for performing semantic similarity calculation on the user input question and the questions of the N knowledge points according to the word vector characteristics, and taking the answer corresponding to the knowledge point with the highest similarity as the answer of the question input by the user.
In a third aspect, an embodiment of the present invention provides a customer service device based on user behavior data, including a processor and a memory, where the processor performs modeling and training of a knowledge point model for implementing any one of the data; the memory stores a knowledge base system, knowledge point identification classification, a model and a program.
According to the customer service method and system based on the user behavior data, the weight division correction method based on the user behavior data is used for supplementing information for knowledge point identification in the customer service processing process, and accuracy of knowledge point classification is improved. Compared with the prior art, in the implementation of the invention, the knowledge point category weight tree is obtained according to the user behavior data and the knowledge base system, the knowledge point category weight tree corresponding to the knowledge point category tree is generated based on the basic data of the user in the service system and the behavior data concerning the product and the service, and the knowledge point category weight tree integrates the behavior information of the product or the service used by the user and can be used for carrying out the subsequent knowledge point identification correction. According to the problems input by the user, word vector characteristic data are obtained, the problems input by user consultation are received, text characteristic data are generated through processing of text error correction, text word segmentation and the like and serve as input of user conversation of customer service processing, the text error correction corrects spelling errors or homophone errors possibly input by the user according to a domain professional dictionary, and word vector characteristic data which accord with the use scene of the professional domain are obtained on the basis of a text word segmentation result and serve as input data of a subsequent knowledge point recognition step. Acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data; the method comprises the steps of carrying out preliminary knowledge point identification and weight correction on input problems in the process of customer service consultation by a user to obtain a knowledge point category, wherein the knowledge point considers product service information concerned by the user, so that the prediction of the knowledge point category really expected to be consulted by the user is more accurate. And matching similar questions with knowledge points belonging to the knowledge point category according to the knowledge point category to obtain answers to the questions input by the user, searching for similar questions according to the corrected knowledge point category and providing answers to the knowledge points, wherein the searched knowledge points are more easily close to the consultation intention of the user based on the more accurate knowledge point category.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating a knowledge point category weight tree acquisition process according to an embodiment of the present invention;
FIG. 3 is a system block diagram of an embodiment of the present invention;
FIG. 4 is a system block diagram of a weight tree generation submodule according to an embodiment of the present invention;
FIG. 5 is a diagram of an apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, those skilled in the art can obtain the embodiments without any inventive step in advance, and the embodiments are within the protection scope of the present invention.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The embodiment of the invention provides a customer service method based on user behavior data, as shown in fig. 1, comprising the following steps:
and S110, acquiring a knowledge point category weight tree according to the user behavior data and the knowledge base system.
Preferably, S110 includes:
s1101, establishing a user behavior model according to user service data collected from a service system.
According to the basic information of the user, such as the integrity and other conditions of the user account, the conditions of the user account associated with a bank card, a social contact account and the like, a basic portrait model of the user can be established based on the basic information so as to represent the account characteristics of the user;
based on the basic information of the user, a data structure containing information such as the account number of the user, the types of products and services purchased and concerned, links, time, times and the like is expanded to serve as a basic portrait model of the user, and the basic portrait model is used for describing information data of the user in the system.
The method comprises the steps of establishing mapping of corresponding knowledge points according to basic information, establishing the attention degree of a user to the category of the knowledge points according to the clicking condition or the maintenance condition of the user to account information, collecting business data of goods purchase or service transaction of the user and data (such as shopping cart addition, frequent clicking rate and the like) of goods or service attention of the user in combination with products and services of a business system, and creating a data updating module which is in butt joint with the business system and executes tasks at fixed time based on the data, wherein the data updating module realizes the collection and updating tasks of the data and is called as a user behavior module to represent the attention degree of the user on the products and the services.
And S1102, acquiring a knowledge point category weight tree corresponding to the structure of the knowledge base system according to the knowledge base system and the user behavior model.
Preferably, S1102, as shown in fig. 2, includes:
s1102-1, constructing a product service knowledge point category mapping tree according to a product service system and a knowledge base system of the business system, wherein the method comprises the following steps:
establishing a mapping relation between a product service system of the business system and the knowledge base system to obtain a product service system tree; the product service system tree comprises different product service classifications, and each product service classification comprises different product service subclasses;
constructing a knowledge point system tree according to the knowledge base system; the knowledge point system tree comprises different knowledge point classifications, and each knowledge point classification comprises different knowledge point subclasses;
constructing a product service knowledge point category mapping tree according to the product service system tree and the knowledge point system tree; the product service knowledge point category mapping tree comprises the knowledge point categories, and each knowledge point category comprises the different product service subclasses.
S1102-2, according to the user behavior model, carrying out quantization processing on the product service knowledge point category mapping tree, and obtaining a quantized product service knowledge point category mapping tree.
S1102-3, constructing a knowledge point category attention tree according to the product service touch degree.
S1102-4, acquiring attention freshness according to attention time of a user to product service.
S1102-5, acquiring a knowledge point category weight tree according to the quantized product service knowledge point category mapping tree, the knowledge point category attention degree tree and the attention degree freshness.
Preferably, step S1102 provides the following embodiments, taking the product service business field as an example, the determining step of the knowledge point category weight tree includes traversing a path of a product or service, and setting a weight for a knowledge point category corresponding to a corresponding product or service according to a product and service that a user has purchased or participated in, and the detailed steps are as follows:
1) and (3) mapping the product service system in the vertical field with the knowledge base system to construct a product service system tree:
Treeproduct＝{ClassA:{P1,P2,P3,...,SA1,SA2,SA3,...},ClassB:{SB1,SB2,SB3,...},...}。
wherein, TreeproductRepresenting a product service architecture Tree, ClassARepresenting a certain category, P, in the product service hierarchyiIndicates a certain product, S, involved in the categoryAiIndicating a certain service involved in the category, ClassBRepresenting a certain category, S, in the product service hierarchyBiIndicating a certain service involved in the category.
2) Constructing a knowledge point system tree:
Treekb＝{ClassX:{K1,K2,K3,..},ClassY:{Ki,Kj,Kk,...},...}。
wherein, TreekbRepresenting a tree of knowledge point systems, ClassXRepresenting a certain class, K, in a tree of knowledge points hierarchyiIndicating a certain knowledge involved in the category, ClassYRepresenting a certain class, K, in a tree of knowledge points hierarchyi，Kj，KkTo representSome knowledge involved in the category.
3) According to the condition that the knowledge points cover the product service, constructing a product service knowledge point category mapping tree:
Treemap＝{ClassX:{P1,P2,P3,..},ClassY:{SA1,SA2,SA3,SB1,SB2,SB3,...},...}
wherein, TreemapClass mapping tree, Class, representing product service knowledge pointsXRepresenting a certain class, P, in a tree of knowledge points hierarchyiRepresenting a product mapped into the knowledge point Class, ClassYRepresenting a certain class, S, in a knowledge point hierarchy treeAiRepresentation mapping to ClassYA certain service in the knowledge point category, SBiRepresentation mapping to ClassYA certain service in the knowledge point category.
4) According to the condition that the knowledge points cover the product service, a product service knowledge point category mapping tree is constructed, for example, data is obtained according to the modeling of the user behavior data, 1 represents that the product service is concerned, and 0 represents that no:
Treehit＝{ClassX:{1,1,0,..},ClassY:{0,1,1,0,1,0,...},...}
wherein, TreehitAnd showing the specific touch condition of the product service knowledge point category mapping tree.
5) Defining a knowledge point category attention tree according to the product service touch degree:
Treeheat＝{ClassX:(1+log(n1)),ClassY:(1+log(n2)),...}
wherein, TreeheatTree-based representation of a knowledge point class TreehitNumber of 1 in the specific category, e.g. niA calculated attention value for each class, e.g. 1+ log (n)i)。
6) An attention freshness of a product service is defined as t (t) 1/(1+ log (t)), where t is a time variable and the attention freshness decays as time increases. And setting a rule, if a product service under a certain Class is newly touched, resetting the value of t to be 1, namely resetting the value of T (t) to be 1, and recovering the freshness of the Class to be 1.
7) Integrating the above attention and freshness to define a demand weight tree for the knowledge point category (where n isiNumber of services served for product of interest in category):
wherein, TreeweightRequirement weight Tree, WeightClass, representing classes of knowledge pointsXRepresenting knowledge point ClassXA weight index of (1) having a value of
And S120, acquiring word vector characteristic data according to the questions input by the user.
And receiving the problem of user consultation input, and generating text characteristic data through processing such as text error correction and text word segmentation to be used as the input of user session of customer service processing.
The text error correction is used for correcting spelling errors or homophone errors possibly input by a user according to a domain professional dictionary, and can be used for correcting wrong words based on an N-Gram combined probabilistic language model.
The text segmentation is based on a user-defined service field dictionary, and various segmentation tools can be adopted, and the embodiment is not particularly limited.
Based on the text segmentation result, a pre-trained Chinese word vector model, such as an open-source Chinese word vector corpus, may be used in the step of generating text feature data, in which each row represents a word and its corresponding low-dimensional dense vector (e.g., 100-dimensional), such as "financing 0.0031460.5826710.049029-0.3128030.5229860.026432-0.0971150.194231-0.362708 … …"
And extracting Word vector characteristics of the Word segmentation sequence, or training Word vectors of the corpus of a knowledge base in the professional field by adopting a Word2vec training tool of genim to obtain a Word vector model according with the use scene of the professional field, and obtaining Word vector characteristic data of the Word segmentation sequence.
S130, acquiring the knowledge point category corresponding to the input question according to the knowledge point category weight tree and the word vector characteristic data.
Training a knowledge point model according to a pre-trained knowledge base corpus to obtain a knowledge point identification model;
judging the knowledge point category to which the knowledge point belongs according to the word vector characteristic data and the knowledge point identification model, and acquiring knowledge point category information; the knowledge point category information comprises various knowledge point categories and corresponding scores;
calculating the difference between the highest-scoring knowledge point category and the next highest-scoring knowledge point category;
if the difference is lower than a category confusion threshold, multiplying the weight of the knowledge point category with the highest score corresponding to the knowledge point category weight tree by the highest score to obtain a first value; multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem;
and if the difference is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question.
Preferably, the following examples are given,
and (3) carrying out knowledge point classification model training on the corpus data of the knowledge base by adopting neural networks such as LSTM or TextCNN and the like to obtain a knowledge point identification model. As shown in fig. 3, knowledge point identification is divided into two steps:
(1) and (3) performing intention identification and judgment according to word vector characteristic data, outputting a knowledge point category result sequence through a Softmax layer, and arranging in a descending order according to Score:
Output＝{(Label1,Score1),(Label2,Score2),(Label3,Score3),...}，
calculating a difference value between the knowledge point category with the highest score and the knowledge point category with the next highest score, comparing the difference value with a category confusion threshold, if the difference value is lower than the category confusion threshold, entering a second step, and if the difference value is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question;
according to the actual situation, a category confusion threshold value is set artificially, for example 0.15, and generally, according to the actual knowledge point classification result, the input problems of wrong knowledge point classification are counted after a period of time, the highest-score knowledge point category and the next-highest-score category are collected, and the difference between the two categories is calculated. This results in some misclassified class difference statistics, and the maximum of these can be taken as the threshold. Thus, if the value is less than the threshold, it is considered that the class correction is required. Of course, after correction, there may be data in which the error that has not been corrected but has been classified is not corrected. Analysis of these data artificially gives a threshold value after the trade-off.
(2) A second step of multiplying the weight of the corresponding knowledge point category with the highest score in the knowledge point category weight tree by the highest score to obtain a first value; and multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem.
S140, according to the knowledge point category, carrying out similar question matching on the knowledge points belonging to the knowledge point category, and obtaining the question answer input by the user.
According to the obtained knowledge point category, similar question matching recall is carried out on the knowledge points under the category, and the function is divided into two steps:
(1) firstly, cosine distance similarity calculation is carried out through TF-IDF characteristics, and the first N knowledge points with the highest similarity are obtained; the method comprises the steps of segmenting words of input sentences, calculating TF-IDF values of the words based on knowledge point linguistic data, generating vectors corresponding to the words, obtaining corresponding vectors of all knowledge points in a knowledge base in the same mode, calculating cosine similarity scores, arranging in a descending mode, and screening out the first N knowledge points.
(2) And secondly, performing Word vector feature extraction on the standard questions of the N knowledge points, further performing semantic similarity calculation by adopting a Word2vec tool of a genim package and Word vector features of the questions input by the user, and outputting knowledge point answers with the highest similarity.
S150, interface transmission processing of information input and answer output in the user conversation process, overtime processing and manual transfer processing in the user conversation process and other flow processing.
An embodiment of the present invention further provides a customer service system based on user behavior data, as shown in fig. 3, including:
the knowledge point category weight tree acquisition module is used for acquiring a knowledge point category weight tree according to the user behavior data and the knowledge base system;
the word vector characteristic acquisition module is used for acquiring word vector characteristic data according to the problems input by the user;
the knowledge point category identification module is used for acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data;
and the similar question matching module is used for matching similar questions with knowledge points belonging to the knowledge point category according to the knowledge point category to obtain the question answers input by the user.
According to an embodiment of the present invention, the system further includes a session management module, configured to perform interface transmission processing for information input and answer output in the user session process, and perform flow processing such as timeout processing and manual transfer processing in the user session process.
The knowledge point category weight tree obtaining module comprises:
the user behavior modeling submodule is used for establishing a user behavior model according to user service data collected from the service system;
and the weight tree obtaining submodule is used for obtaining the knowledge point category weight tree corresponding to the structure of the knowledge base system according to the knowledge base system and the user behavior model.
The weight tree generation submodule, as shown in fig. 4, includes:
the product service knowledge point category mapping tree acquisition submodule is used for constructing a product service knowledge point category mapping tree according to a product service system and a knowledge base system of the service system;
the product service knowledge point category mapping tree quantization submodule is used for carrying out quantization processing on the product service knowledge point category mapping tree according to the user behavior model to obtain a quantized product service knowledge point category mapping tree;
the attention tree obtaining submodule is used for constructing a knowledge point category attention tree according to the product service touch;
the concerned freshness obtaining submodule is used for obtaining concerned freshness according to the concerned time of the user on the product service;
and the weight tree construction submodule is used for acquiring a knowledge point category weight tree according to the quantized product service knowledge point category mapping tree, the knowledge point category attention tree and the attention freshness.
The product service knowledge point category mapping tree obtaining submodule comprises:
the product service system tree construction submodule is used for establishing a mapping relation between a product service system of the business system and the knowledge base system to obtain a product service system tree; the product service system tree comprises different product service classifications, and each product service classification comprises different product service subclasses;
the knowledge point system tree construction submodule is used for constructing a knowledge point system tree according to the knowledge base system; the knowledge point system tree comprises different knowledge point classifications, and each knowledge point classification comprises different knowledge point subclasses;
the mapping tree construction submodule is used for constructing a product service knowledge point category mapping tree according to the product service system tree and the knowledge point system tree; the product service knowledge point category mapping tree comprises the knowledge point categories, and each knowledge point category comprises the different product service subclasses.
The knowledge point category identification module comprises:
the knowledge point model construction submodule is used for carrying out knowledge point model training according to the corpus of a pre-trained knowledge base to obtain a knowledge point identification model;
the knowledge point category information acquisition submodule is used for judging the category of the knowledge point to which the knowledge point belongs according to the word vector characteristic data and the knowledge point identification model and acquiring the category information of the knowledge point; the knowledge point category information comprises various knowledge point categories and corresponding scores; the knowledge point category identification submodule is used for calculating the difference between the knowledge point category with the highest score and the knowledge point category with the next highest score; if the difference is lower than a category confusion threshold, multiplying the weight of the knowledge point category with the highest score corresponding to the knowledge point category weight tree by the highest score to obtain a first value; multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem; and if the difference is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question.
The similarity problem matching module comprises:
the cosine distance similarity calculation operator module is used for calculating the cosine distance similarity of the knowledge points belonging to the knowledge point category according to the TF-IDF characteristics to obtain the first N knowledge points with the highest similarity;
and the semantic similarity calculation operator module is used for performing semantic similarity calculation on the user input question and the questions of the N knowledge points according to the word vector characteristics, and taking the answer corresponding to the knowledge point with the highest similarity as the answer of the question input by the user.
The embodiment of the invention also provides a customer service device based on the user behavior data, which comprises a processor and a memory, wherein the processor is used for implementing any data modeling and knowledge point model training; the memory stores a knowledge base system, knowledge point identification classification, a model and a program.
As shown in fig. 5, the memory includes a storage medium ROM and a storage medium RAM, the memory is connected to a system bus, the processor is connected to the system bus, and the system bus is connected to the network.
According to the customer service method and system based on the user behavior data, the weight division correction method based on the user behavior data is used for supplementing information for knowledge point identification in the customer service processing process, and accuracy of knowledge point classification is improved. Compared with the prior art, in the implementation of the invention, the knowledge point category weight tree is obtained according to the user behavior data and the knowledge base system, the knowledge point category weight tree corresponding to the knowledge point category tree is generated based on the basic data of the user in the service system and the behavior data concerning the product and the service, and the knowledge point category weight tree integrates the behavior information of the product or the service used by the user and can be used for carrying out the subsequent knowledge point identification correction. According to the problems input by the user, word vector characteristic data are obtained, the problems input by user consultation are received, text characteristic data are generated through processing of text error correction, text word segmentation and the like and serve as input of user conversation of customer service processing, the text error correction corrects spelling errors or homophone errors possibly input by the user according to a domain professional dictionary, and word vector characteristic data which accord with the use scene of the professional domain are obtained on the basis of a text word segmentation result and serve as input data of a subsequent knowledge point recognition step. Acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data; the method comprises the steps of carrying out preliminary knowledge point identification and weight correction on input problems in the process of customer service consultation by a user to obtain a knowledge point category, wherein the knowledge point considers product service information concerned by the user, so that the prediction of the knowledge point category really expected to be consulted by the user is more accurate. And matching similar questions with knowledge points belonging to the knowledge point category according to the knowledge point category to obtain answers to the questions input by the user, searching for similar questions according to the corrected knowledge point category and providing answers to the knowledge points, wherein the searched knowledge points are more easily close to the consultation intention of the user based on the more accurate knowledge point category.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus embodiment, since it is substantially similar to the method embodiment, it is relatively simple to describe, and reference may be made to some descriptions of the method embodiment for relevant points. Those skilled in the art will appreciate that the modules in the devices in the embodiments may be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components of the embodiments may be combined into one module or unit or component, and furthermore they may be divided into a plurality of sub-modules or sub-units or sub-components. All of the features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or elements of any method or apparatus so disclosed, may be combined in any combination, except combinations where at least some of such features and/or processes or elements are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (12)
1. A customer service method based on user behavior data is characterized by comprising the following steps:
acquiring a knowledge point category weight tree according to user behavior data and a knowledge base system;
acquiring word vector characteristic data according to a problem input by a user;
acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data;
and according to the knowledge point category, carrying out similar question matching on the knowledge points belonging to the knowledge point category to obtain the question answers input by the user.
2. The method of claim 1, wherein obtaining the knowledge point category weight tree according to the user behavior data and the knowledge base system comprises:
establishing a user behavior model according to user service data collected from a service system;
and acquiring a knowledge point category weight tree corresponding to the structure of the knowledge base system according to the knowledge base system and the user behavior model.
3. The method of claim 2, wherein obtaining the knowledge point category weight tree corresponding to the knowledge base architecture according to the knowledge base architecture and the user behavior model comprises:
constructing a product service knowledge point category mapping tree according to a product service system and a knowledge base system of the service system;
according to the user behavior model, carrying out quantization processing on the product service knowledge point category mapping tree to obtain a quantized product service knowledge point category mapping tree;
constructing a knowledge point category attention tree according to the product service touch;
acquiring attention freshness according to attention time of a user to product service;
and acquiring a knowledge point category weight tree according to the quantized product service knowledge point category mapping tree, the knowledge point category attention tree and the attention freshness.
4. The method of claim 3, wherein constructing a product service knowledge point category mapping tree according to a product service architecture and a knowledge base architecture of the business system comprises:
establishing a mapping relation between a product service system of the business system and the knowledge base system to obtain a product service system tree; the product service system tree comprises different product service classifications, and each product service classification comprises different product service subclasses;
constructing a knowledge point system tree according to the knowledge base system; the knowledge point system tree comprises different knowledge point classifications, and each knowledge point classification comprises different knowledge point subclasses;
constructing a product service knowledge point category mapping tree according to the product service system tree and the knowledge point system tree; the product service knowledge point category mapping tree comprises the knowledge point categories, and each knowledge point category comprises the different product service subclasses.
5. The method according to claim 1, wherein the obtaining the knowledge point category corresponding to the input question according to the knowledge point category weight tree and the word vector feature data comprises:
training a knowledge point model according to a pre-trained knowledge base corpus to obtain a knowledge point identification model;
judging the knowledge point category to which the knowledge point belongs according to the word vector characteristic data and the knowledge point identification model, and acquiring knowledge point category information; the knowledge point category information comprises various knowledge point categories and corresponding scores;
calculating the difference between the highest-scoring knowledge point category and the next highest-scoring knowledge point category;
if the difference is lower than a category confusion threshold, multiplying the weight of the knowledge point category with the highest score corresponding to the knowledge point category weight tree by the highest score to obtain a first value; multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem;
and if the difference is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question.
6. The method according to claim 1, wherein the performing similar question matching on the knowledge points belonging to the knowledge point category according to the knowledge point category to obtain the answer to the question input by the user comprises:
calculating the cosine distance similarity of the knowledge points belonging to the knowledge point category according to TF-IDF characteristics to obtain the first N knowledge points with the highest similarity;
and performing semantic similarity calculation on the questions input by the user and the questions of the N knowledge points according to the word vector characteristics, and taking the answer corresponding to the knowledge point with the highest similarity as the answer of the questions input by the user.
7. A customer service system based on user behavior data, comprising:
the knowledge point category weight tree acquisition module is used for acquiring a knowledge point category weight tree according to the user behavior data and the knowledge base system;
the word vector characteristic acquisition module is used for acquiring word vector characteristic data according to the problems input by the user;
the knowledge point category identification module is used for acquiring the knowledge point category corresponding to the input problem according to the knowledge point category weight tree and the word vector characteristic data;
and the similar question matching module is used for matching similar questions with knowledge points belonging to the knowledge point category according to the knowledge point category to obtain the question answers input by the user.
8. The system according to claim 7, wherein the knowledge point category weight tree obtaining module comprises:
the user behavior modeling submodule is used for establishing a user behavior model according to user service data collected from the service system;
and the weight tree obtaining submodule is used for obtaining the knowledge point category weight tree corresponding to the structure of the knowledge base system according to the knowledge base system and the user behavior model.
9. The system of claim 8, wherein the weight tree generation submodule comprises:
the product service knowledge point category mapping tree acquisition submodule is used for constructing a product service knowledge point category mapping tree according to a product service system and a knowledge base system of the service system;
the product service knowledge point category mapping tree quantization submodule is used for carrying out quantization processing on the product service knowledge point category mapping tree according to the user behavior model to obtain a quantized product service knowledge point category mapping tree;
the attention tree obtaining submodule is used for constructing a knowledge point category attention tree according to the product service touch;
the concerned freshness obtaining submodule is used for obtaining concerned freshness according to the concerned time of the user on the product service;
and the weight tree construction submodule is used for acquiring a knowledge point category weight tree according to the quantized product service knowledge point category mapping tree, the knowledge point category attention tree and the attention freshness.
10. The system according to claim 9, wherein the product service knowledge point category mapping tree obtaining sub-module comprises:
the product service system tree construction submodule is used for establishing a mapping relation between a product service system of the business system and the knowledge base system to obtain a product service system tree; the product service system tree comprises different product service classifications, and each product service classification comprises different product service subclasses;
the knowledge point system tree construction submodule is used for constructing a knowledge point system tree according to the knowledge base system; the knowledge point system tree comprises different knowledge point classifications, and each knowledge point classification comprises different knowledge point subclasses;
the mapping tree construction submodule is used for constructing a product service knowledge point category mapping tree according to the product service system tree and the knowledge point system tree; the product service knowledge point category mapping tree comprises the knowledge point categories, and each knowledge point category comprises the different product service subclasses.
11. The system of claim 7, wherein the knowledge point class identification module comprises:
the knowledge point model construction submodule is used for carrying out knowledge point model training according to the corpus of a pre-trained knowledge base to obtain a knowledge point identification model;
the knowledge point category information acquisition submodule is used for judging the category of the knowledge point to which the knowledge point belongs according to the word vector characteristic data and the knowledge point identification model and acquiring the category information of the knowledge point; the knowledge point category information comprises various knowledge point categories and corresponding scores;
the knowledge point category identification submodule is used for calculating the difference between the knowledge point category with the highest score and the knowledge point category with the next highest score; if the difference is lower than a category confusion threshold, multiplying the weight of the knowledge point category with the highest score corresponding to the knowledge point category weight tree by the highest score to obtain a first value; multiplying the weight of the knowledge point category with the second highest score corresponding to the knowledge point category weight tree by the second highest score to obtain a second value, and taking the knowledge point category corresponding to the larger value of the first value and the second value as the knowledge point category corresponding to the input problem; and if the difference is not lower than the category confusion threshold, taking the knowledge point category with the highest score as the knowledge point category corresponding to the input question.
12. The system of claim 7, wherein the affinity problem matching module comprises:
the cosine distance similarity calculation operator module is used for calculating the cosine distance similarity of the knowledge points belonging to the knowledge point category according to the TF-IDF characteristics to obtain the first N knowledge points with the highest similarity;
and the semantic similarity calculation operator module is used for performing semantic similarity calculation on the user input question and the questions of the N knowledge points according to the word vector characteristics, and taking the answer corresponding to the knowledge point with the highest similarity as the answer of the question input by the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911365582.2A CN111143533B (en) | 2019-12-26 | 2019-12-26 | Customer service method and system based on user behavior data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911365582.2A CN111143533B (en) | 2019-12-26 | 2019-12-26 | Customer service method and system based on user behavior data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111143533A true CN111143533A (en) | 2020-05-12 |
| CN111143533B CN111143533B (en) | 2023-06-30 |
Family
ID=70520441
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911365582.2A Active CN111143533B (en) | 2019-12-26 | 2019-12-26 | Customer service method and system based on user behavior data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111143533B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111782782A (en) * | 2020-06-09 | 2020-10-16 | 苏宁金融科技(南京)有限公司 | Consultation reply method and device for intelligent customer service, computer equipment and storage medium |
| CN111881274A (en) * | 2020-07-13 | 2020-11-03 | 北京捷通华声科技股份有限公司 | Method, device and processor for determining answer of question |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109933779A (en) * | 2017-12-18 | 2019-06-25 | 苏宁云商集团股份有限公司 | User's intension recognizing method and system |
| CN110532400A (en) * | 2019-09-04 | 2019-12-03 | 江苏苏宁银行股份有限公司 | Knowledge base maintenance method and device based on text classification prediction |
-
2019
- 2019-12-26 CN CN201911365582.2A patent/CN111143533B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109933779A (en) * | 2017-12-18 | 2019-06-25 | 苏宁云商集团股份有限公司 | User's intension recognizing method and system |
| CN110532400A (en) * | 2019-09-04 | 2019-12-03 | 江苏苏宁银行股份有限公司 | Knowledge base maintenance method and device based on text classification prediction |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111782782A (en) * | 2020-06-09 | 2020-10-16 | 苏宁金融科技(南京)有限公司 | Consultation reply method and device for intelligent customer service, computer equipment and storage medium |
| CN111881274A (en) * | 2020-07-13 | 2020-11-03 | 北京捷通华声科技股份有限公司 | Method, device and processor for determining answer of question |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111143533B (en) | 2023-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108520343B (en) | Risk model training method, risk identification device, risk identification equipment and risk identification medium | |
| CN111159407B (en) | Method, apparatus, device and medium for training entity recognition and relation classification model | |
| CN111199474A (en) | Risk prediction method and device based on network diagram data of two parties and electronic equipment | |
| CN111797320B (en) | Data processing method, device, equipment and storage medium | |
| CN110019822B (en) | Few-sample relation classification method and system | |
| CN113177700B (en) | Risk assessment method, system, electronic equipment and storage medium | |
| US11811708B2 (en) | Systems and methods for generating dynamic conversational responses using cluster-level collaborative filtering matrices | |
| CN112819024B (en) | Model processing method, user data processing method and device and computer equipment | |
| CN111143533B (en) | Customer service method and system based on user behavior data | |
| CN115130711A (en) | Data processing method and device, computer and readable storage medium | |
| CN111625715A (en) | Information extraction method and device, electronic equipment and storage medium | |
| CN110489423B (en) | Information extraction method and device, storage medium and electronic equipment | |
| CN115545886A (en) | Overdue risk identification method, overdue risk identification device, overdue risk identification equipment and storage medium | |
| CN111582912A (en) | Portrait modeling method based on deep embedding clustering algorithm | |
| CN111210332A (en) | Method and device for generating post-loan management strategy and electronic equipment | |
| CN112836750A (en) | System resource allocation method, device and equipment | |
| CN115794898B (en) | Financial information recommendation method and device, electronic equipment and storage medium | |
| US20140324524A1 (en) | Evolving a capped customer linkage model using genetic models | |
| US20140324523A1 (en) | Missing String Compensation In Capped Customer Linkage Model | |
| CN114529399A (en) | User data processing method, device, computer equipment and storage medium | |
| CN111382265B (en) | Searching method, device, equipment and medium | |
| CN112991025A (en) | Intelligent insurance recommendation method, system and equipment and computer readable storage medium | |
| CN113571198A (en) | Conversion rate prediction method, device, equipment and storage medium | |
| CN115203382A (en) | Service problem scene identification method and device, electronic equipment and storage medium | |
| CN114036267A (en) | Conversation method and system |
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 |