WO2019223058A1 - 提高智能客服应答率的方法、设备、存储介质及装置 - Google Patents

提高智能客服应答率的方法、设备、存储介质及装置 Download PDF

Info

Publication number
WO2019223058A1
WO2019223058A1 PCT/CN2018/092989 CN2018092989W WO2019223058A1 WO 2019223058 A1 WO2019223058 A1 WO 2019223058A1 CN 2018092989 W CN2018092989 W CN 2018092989W WO 2019223058 A1 WO2019223058 A1 WO 2019223058A1
Authority
WO
WIPO (PCT)
Prior art keywords
customer service
target standard
intelligent customer
question
knowledge
Prior art date
Application number
PCT/CN2018/092989
Other languages
English (en)
French (fr)
Inventor
于凤英
王健宗
肖京
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2019223058A1 publication Critical patent/WO2019223058A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/332Query formulation
    • G06F16/3329Natural language query formulation or dialogue systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION 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
    • G06Q30/00Commerce
    • G06Q30/02Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201Market modelling; Market analysis; Collecting market data
    • G06Q30/0203Market surveys; Market polls
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • G06F40/35Discourse or dialogue representation

Definitions

  • the present application relates to the technical field of intelligent customer service, and in particular, to a method, a device, a storage medium, and a device for improving the response rate of intelligent customer service.
  • the main purpose of this application is to provide a method, device, storage medium and device for improving the response rate of intelligent customer service, which aims to solve the technical problem of low response rate of intelligent customer service in the prior art.
  • the present application provides a method for improving the response rate of intelligent customer service.
  • the method for increasing the response rate of intelligent customer service includes the following steps:
  • the present application also proposes a device for improving the response rate of the intelligent customer service.
  • the device for improving the response rate of the intelligent customer service includes a memory, a processor, and the memory and the processor.
  • the readable instructions for improving the response rate of the intelligent customer service are run, and the readable instructions for improving the response rate of the intelligent customer service are configured as steps of the method for improving the response rate of the intelligent customer service as described above.
  • the present application also proposes a storage medium storing a readable instruction for improving the response rate of the intelligent customer service, which is implemented when the readable instruction for improving the response rate of the intelligent customer service is executed by the processor. Steps of the method for improving the response rate of the intelligent customer service as described above.
  • the present application also proposes a device for improving the response rate of intelligent customer service.
  • the device for improving the response rate of intelligent customer service includes: a classification module, a calculation module, a clustering module, an acquisition module, and an addition module;
  • the classification module is configured to obtain unanswered questions from the intelligent customer service, classify the unanswered questions, and obtain knowledge-based questions;
  • the calculation module is configured to calculate a first similarity between the knowledge-type problems
  • the clustering module is configured to cluster each knowledge-type problem by an affinity propagation clustering algorithm according to the first similarity to obtain a plurality of clustering clusters;
  • the acquisition module is configured to use a cluster center of each cluster cluster as a target standard problem, and obtain a target standard response corresponding to the target standard problem;
  • the adding module is configured to add the target standard question and the corresponding target standard response to the knowledge base of the intelligent customer service.
  • FIG. 1 is a schematic structural diagram of a device for improving a response rate of a smart customer service in a hardware operating environment according to a solution of an embodiment of the present application
  • FIG. 2 is a schematic flowchart of a first embodiment of a method for improving a response rate of a smart customer service in this application;
  • FIG. 3 is a schematic flowchart of a second embodiment of a method for improving a response rate of an intelligent customer service in this application;
  • FIG. 4 is a schematic flowchart of a third embodiment of a method for improving a response rate of a smart customer service in this application;
  • FIG. 5 is a structural block diagram of a first embodiment of an apparatus for improving a response rate of an intelligent customer service in this application.
  • FIG. 1 is a schematic structural diagram of a device for improving a response rate of a smart customer service in a hardware operating environment according to an embodiment of the present application.
  • the device for improving the response rate of intelligent customer service may include: a processor 1001, such as a central processing unit (Central Processing Unit (CPU), communication bus 1002, user interface 1003, network interface 1004, and memory 1005.
  • the communication bus 1002 is configured to implement connection and communication between these components.
  • the user interface 1003 may include a display screen.
  • the optional user interface 1003 may further include a standard wired interface and a wireless interface.
  • the wired interface of the user interface 1003 may be a USB interface in this application.
  • the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WIreless-FIdelity (WI-FI) interface).
  • the memory 1005 may be a high-speed random access memory (Random Access Memory (RAM) memory or non-volatile memory Memory (NVM), such as disk storage.
  • the memory 1005 may optionally be a storage device independent of the foregoing processor 1001.
  • FIG. 1 does not constitute a limitation on the device for improving the response rate of the intelligent customer service, and may include more or fewer components than shown in the figure, or combine some components, or different Parts arrangement.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and readable instructions for improving the response rate of the intelligent customer service.
  • the network interface 1004 is mainly configured to connect to a background server for data communication with the background server;
  • the user interface 1003 is mainly configured to connect to an intelligent customer service device;
  • the response rate device calls the readable instructions stored in the memory 1005 to improve the response rate of the intelligent customer service through the processor 1001, and executes the method for improving the response rate of the intelligent customer service provided by the embodiment of the present application.
  • FIG. 2 is a schematic flowchart of a first embodiment of a method for improving a response rate of a smart customer service in this application, and a first embodiment of a method for improving a response rate of a smart customer service in this application is proposed.
  • the method for improving the response rate of intelligent customer service includes the following steps:
  • Step S10 Obtain unanswered questions from the intelligent customer service, classify the unanswered questions, and obtain knowledge-based questions.
  • the execution subject of this embodiment is a device for improving the response rate of intelligent customer service, wherein the device for increasing the response rate of intelligent customer service may be an electronic device such as a personal computer or a server.
  • Unanswered questions are divided into two categories, one is a knowledge-based problem, and the other is a non-knowledge-type problem.
  • Knowledge-based questions can usually find corresponding standard answers, which can be added to the knowledge base.
  • Non-knowledge-type issues such as emotional and emotional issues, do not have standard responses, no additional considerations are needed, and there is no need to add to the knowledge base. Therefore, after classification and processing, we retain the knowledge-based questions in the unanswered questions in order to add subsequent processing to the knowledge-based questions.
  • a vocabulary can be constructed, and the vocabulary can be regarded as a huge set. All words in the training set text of the knowledge-based questions are stored in the vocabulary. In addition, the number, index, and type of each word can also be calculated during the construction process, and the types include corpora and tags. In the word vector space, these knowledge-based problems have similar word vectors, and these approximate word vectors correspond to the same classification labels. During the iterative training process, this correlation will continue to propagate until an accurate classification model is trained. Then, the unanswered question can be classified by the classification model to obtain the knowledge-type question.
  • Step S20 Calculate a first similarity between the knowledge-based problems.
  • word segmentation can be performed for each knowledge-type problem, the TF-IDF value of the word is calculated as the word feature, each knowledge-type problem is represented as a word vector, and the cosine distance between word vectors is calculated as the distance between each knowledge-type problem First similarity measure.
  • TF-IDF is actually: TF * IDF, TF Term Frequency, IDF Inverse Document Frequency Frequency).
  • TF indicates how often an entry appears in document d.
  • the main idea of IDF is: if there are fewer documents containing the entry t, the larger the IDF, it means that the entry t has a good ability to distinguish categories.
  • calculating the first similarity between the knowledge-type problems specifically includes: performing word segmentation processing on each knowledge-type problem to obtain all the words in the knowledge-type problem, and calculating the TF of the words -IDF value; representing each knowledge-based problem as a word vector composed of words and TF-IDF values of words; calculating the cosine distance between each word vector, and using the cosine distance as the first A similarity.
  • Step S30 cluster each knowledge-type problem by an affinity propagation clustering algorithm according to the first similarity to obtain a plurality of clusters.
  • affinity propagation clustering Affinity The Propagation (AP) algorithm performs clustering based on the similarity between data points, which can be symmetric or asymmetric.
  • the affinity propagation clustering algorithm does not need to determine the number of clusters first, but treats all data points as cluster centers in a potential sense. According to each first similarity measure, a similarity matrix can be constructed, with each knowledge-type problem as a node, and the first similarity measure between each knowledge-type problem as the value of the matrix, and then clustered by the AP algorithm to obtain multiple clusters. Class cluster.
  • Step S40 Use the cluster center of each cluster as the target standard problem, and obtain a target standard response corresponding to the target standard problem.
  • a cluster center may be obtained from the cluster cluster, and the cluster center is most similar to other problems in the corresponding cluster cluster and has the most representativeness of the cluster cluster. Therefore, The clustering center is selected as the target standard problem of the knowledge-type problem, and other problems in the clustering cluster can be used as the extended problems corresponding to the target standard problem.
  • the corresponding target standard responses can usually be found from the existing database, and the existing database can be queried according to the target standard questions, and the questions related to the target standards can be found. Match the information and use it as a target response.
  • manual intervention may be performed, and the matched data is sorted through the manual customer service to generate a target standard response.
  • Step S50 Add the target standard question and the corresponding target standard response to the knowledge base of the intelligent customer service.
  • the most representative cluster center in each cluster is used as the target standard question, and the target standard question and the corresponding target standard reply are added to the knowledge base of the intelligent customer service, so that there is no response Representative questions can be added to the intelligent customer service knowledge base.
  • the unanswered questions of the intelligent customer service are obtained, the unanswered questions are classified, and the knowledge-based questions are obtained, so that the non-knowledge-type questions obtained by the classification are excluded, and the non-knowledge-type questions do not have standard responses.
  • FIG. 3 is a schematic flowchart of a second embodiment of a method for improving a response rate of a smart customer service according to the present application. Based on the first embodiment shown in FIG. 2 above, a second implementation of a method for improving a response rate of a smart customer service is proposed. example.
  • the step S10 includes:
  • Step S101 Obtain unanswered questions from the intelligent customer service, calculate the frequency of the unanswered questions, and treat the unanswered questions whose frequency exceeds the first preset threshold as pending questions.
  • the intelligent customer service unanswered questions are obtained, and the number of times each unanswered question occurs is counted, so as to obtain the frequency of each unanswered question.
  • the low-frequency unanswered question belongs to the long-tail problem, and the quality of the question is usually low and does not For reference, usually this kind of problem, we do not need to analyze. Therefore, the first preset threshold may be set in advance, and unanswered questions that exceed the first preset threshold are issues that the customer is more concerned about. Such questions are often referenced.
  • the pending problem is a knowledge-based problem, a corresponding standard answer can be found and added to the knowledge base. If the customer asks a related question next time, the intelligent customer service can find the corresponding standard answer from the knowledge base. Answer.
  • Step S102 classify the problem to be processed through a preset classification model to obtain a knowledge-based problem.
  • the preset classification model includes a FastText classification model
  • the FastText classification model includes three parts: a model architecture, a layered classifier (Softmax), and a Chinese language model (N-gram) feature.
  • Softmax is based on Huffman coding and encodes labels, which can greatly reduce the number of model prediction targets, that is, greatly reduce the preset classification model to predict knowledge-based problems in the to-be-processed problems.
  • Quantity FastText N-gram features are added to take local word order into account, thereby achieving more accurate classification of the problem to be processed.
  • step S40 the method further includes:
  • Step S401 Calculate a second similarity between the target standard question and the original standard question in the knowledge base of the intelligent customer service.
  • each original standard question in the knowledge base can be segmented, the TF-IDF value of the word can be calculated, and each original standard question can be expressed as a word vector composed of words and the TF-IDF value of the word, and each original standard can be calculated.
  • the cosine distance between the word vector corresponding to the question and the word vector corresponding to the target standard question, and the cosine distance is used as the second similarity.
  • Step S402 Obtain a maximum value in the second similarity, and determine whether the maximum value exceeds a second preset threshold.
  • the maximum value of the second similarity is obtained, and the original standard problem corresponding to the maximum value is the closest to the target standard problem, and it is determined whether the maximum value exceeds a second preset threshold, and if it exceeds The second preset threshold value indicates that the target standard question has a high correlation with the original standard question corresponding to the maximum value in the knowledge base.
  • Step S403 if the maximum value exceeds the second preset threshold, obtain an original standard question corresponding to the maximum value as a question to be supplemented.
  • the maximum value exceeds the second preset threshold value, for example, the second preset threshold value is 80%, and if the maximum value exceeds 81%, the maximum value is considered to correspond to
  • the original standard question has a high correlation with the target standard question, and may even be the same problem in essence, but there is a difference in expression. It may be proposed to merge the target standard question with the original standard question corresponding to the maximum value. Taking the original standard question corresponding to the maximum value as a to-be-added question, the target standard question is added to the knowledge base as an extended question of the to-be-added question.
  • step S50 specifically includes:
  • Step S501 Add the target standard question and the corresponding target standard reply to the extended question list of the to-be-added questions in the knowledge base of the intelligent customer service.
  • the maximum value in the second similarity exceeds the second preset threshold, it indicates that the original standard problem corresponding to the maximum value has a high correlation with the target standard problem, then the The target standard question and the corresponding target standard reply are added to the extended question list of the to-be-added questions in the knowledge base of the intelligent customer service, and when the user asks the to-be-added question or the extended question of the to-be-added question
  • the intelligent customer service can also display other questions in the extended question list of the to-be-added question, so as to provide users with more comprehensive information about the question they want to consult.
  • the unanswered questions whose frequency exceeds a first preset threshold are regarded as pending questions, so that the pending questions are classified to obtain knowledge questions, and Low frequency non-response questions are eliminated, reducing the workload of subsequent determination of target standard problems and improving the quality of target standard problems supplemented by the knowledge base; calculating the target standard problems and the original standards in the knowledge base of the intelligent customer service
  • FIG. 4 is a schematic flowchart of a third embodiment of a method for improving a response rate of a smart customer service based on the second embodiment shown in FIG. 3 above. example.
  • step S402 the method further includes:
  • Step S404 If the maximum value does not exceed the second preset threshold, use knowledge-based questions other than the target standard question in each cluster as the first extended question.
  • the target standard question is added to the knowledge base in the form of an independent question.
  • Each question in the knowledge base is stored in the form of the original standard question and the corresponding extended question list, so that when a user asks a certain standard question or an issue in the extended question list, the intelligent customer service can The original standard questions and other questions in the corresponding extended question list are also displayed, so as to provide users with more comprehensive information about the questions they want to consult.
  • the target standard problem can be added to the knowledge base as an independent problem, and then other knowledge-based problems in the clustering cluster corresponding to the target standard problem other than the target standard problem are used as the first question.
  • An extended question, the first extended question may be a question in the extended question list of the target standard question.
  • Step S405 Calculate a third similarity between the first extended question and the corresponding target standard question.
  • a third similarity between the first extended problem and the corresponding target standard problem may be calculated, and the first extended problem with a larger value of the third similarity is used as the extended problem list of the target standard problem The problem.
  • the step S405 includes: calculating a statistical feature, a semantic feature, and a topic feature between the first extended question and a corresponding target standard question; and using a logistic regression to combine the statistical feature and the semantic feature Aggregate with the subject feature to obtain a third similarity between the first extended question and the corresponding target standard question.
  • the statistical characteristics include: word co-occurrence rate, TF-IDF value, edit distance, and longest common substring.
  • the word co-occurrence rate, the TF-IDF value, the edit distance, and the longest common substring between the first extended question and the corresponding target standard question may be calculated as the statistical feature.
  • Short-Term Memory constructs a vector of the first extension problem and the target standard problem, calculates a cosine similarity between the vectors, and uses the cosine similarity as the semantic feature.
  • Generate a model from a document theme (Latent Dirichlet Allocation (LDA) generates corresponding topic features for the first extended problem and the corresponding target standard problem.
  • LDA Document Dirichlet Allocation
  • the Logistic Regression (Regression, LR) Based on linear regression, a logical function is applied, and the statistical feature, the semantic feature, and the topic feature can be aggregated through the logistic regression to obtain the first extended problem and corresponding The third similarity between the target standard questions.
  • Step S406 Select N first extended questions from the first extended questions in the order of the third similarity from large to small as the extended list of the target standard questions, where N is greater than or equal to 1 Integer.
  • the third extended degree of similarity can generally be included in the first extended problem in the order of large to small.
  • the step S50 specifically includes:
  • Step S502 Add the extended question list of the target standard question, the target standard question and the corresponding target standard reply to the intelligent customer service knowledge base in the form of independent questions.
  • the maximum value in the second similarity does not exceed the second preset threshold, indicating that the original standard question corresponding to the maximum value is not highly relevant to the target standard question, then the The extended question list of the target standard question, the target standard question and the corresponding target standard reply are added to the intelligent customer service knowledge base in the form of independent questions, and the next time the user asks questions related to the target standard question, The intelligent customer service can find the target standard question and the corresponding extended question list from the knowledge base, thereby realizing the response to the user.
  • the knowledge-type problem other than the target standard problem in each cluster is used as the first expansion problem, and the first calculation is performed.
  • a third similarity between the extended problem and the corresponding target standard problem and selecting the N first extended problems from the first extended problem as the target standard problem according to the third similarity in descending order Extended question list, adding the extended question list of the target standard question, the target standard question and the corresponding target standard reply to the intelligent customer service knowledge base in the form of independent questions, so that the user will inquire the next time
  • the intelligent customer service can find the target standard problem and the corresponding extended problem list from the knowledge base, thereby realizing the response to the user and improving the user experience.
  • an embodiment of the present application further provides a storage medium storing readable instructions for improving the response rate of the intelligent customer service.
  • the readable instructions for improving the response rate of the intelligent customer service are implemented by the processor as described above. The steps of the method to improve the response rate of intelligent customer service are described.
  • the storage medium may be a non-volatile readable storage medium.
  • an embodiment of the present application further provides a device for improving the response rate of intelligent customer service.
  • the device for improving the response rate of intelligent customer service includes: a classification module 10, a calculation module 20, a clustering module 30, an acquisition module 40, Add module 50;
  • the classification module 10 is configured to obtain an unanswered question from a smart customer service, classify the unanswered question, and obtain a knowledge-based question;
  • the calculation module 20 is configured to calculate a first similarity between knowledge-based problems
  • the clustering module 30 is configured to cluster each knowledge-type problem by an affinity propagation clustering algorithm according to the first similarity to obtain multiple clusters;
  • the obtaining module 40 is configured to use the cluster center of each cluster cluster as a target standard problem, and obtain a target standard response corresponding to the target standard problem;
  • the adding module 50 is configured to add the target standard question and the corresponding target standard response to the knowledge base of the intelligent customer service.
  • the method of the embodiment can be implemented by means of software plus a necessary universal hardware platform. Hardware, but in many cases the former is a better implementation.
  • the technical solution of the present application in essence or a part that contributes to the existing technology may be in the form of a software product.
  • the computer software product is stored in a storage medium (such as a Read Only Memory image (ROM) / Random Access Memory (Random Access Memory (RAM), magnetic disks, and optical disks) include a number of instructions for causing a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the embodiments of this application.
  • ROM Read Only Memory image
  • RAM Random Access Memory
  • magnetic disks magnetic disks
  • optical disks include a number of instructions for causing a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the embodiments of this application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Mathematical Physics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Human Computer Interaction (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Databases & Information Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请公开了一种提高智能客服应答率的方法、设备、存储介质及装置,该方法包括:获取智能客服的未应答问题,对未应答问题进行分类,获得知识型问题;计算各知识型问题之间的第一相似度;根据第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;将各聚类簇的聚类中心作为目标标准问题,并获取与目标标准问题对应的目标标准回复;将目标标准问题及对应的目标标准回复添加至智能客服的知识库中。

Description

提高智能客服应答率的方法、设备、存储介质及装置
本申请要求于2018年05月22日提交中国专利局、申请号为201810499843.9、发明名称为“提高智能客服应答率的方法、设备、存储介质及装置”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及智能客服技术领域,尤其涉及一种提高智能客服应答率的方法、设备、存储介质及装置。
背景技术
随着科技的发展,智能客服系统越来越得到重视,但智能客服目前仍不能完全替代人工客服,客户提出的很多问题智能客服无法回答,使得需要投入大量的人工客服对这些问题进行解答,导致人员成本高,用户在使用智能客服进行问询时,经常得不到标准回复,又转接人工服务,导致用户体验差,因此,如何提高智能客服的应答率是亟待解决的技术问题。
上述内容仅设置为辅助理解本申请的技术方案,并不代表承认上述内容是现有技术。
发明内容
本申请的主要目的在于提供一种提高智能客服应答率的方法、设备、存储介质及装置,旨在解决现有技术中智能客服的应答率低的技术问题。
为实现上述目的,本申请提供一种提高智能客服应答率的方法,所述提高智能客服应答率的方法包括以下步骤:
获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
计算各知识型问题之间的第一相似度;
根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
此外,为实现上述目的,本申请还提出一种提高智能客服应答率的设备,所述提高智能客服应答率的设备包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的提高智能客服应答率的可读指令,所述提高智能客服应答率的可读指令配置为实现如上文所述的提高智能客服应答率的方法的步骤。
此外,为实现上述目的,本申请还提出一种存储介质,所述存储介质上存储有提高智能客服应答率的可读指令,所述提高智能客服应答率的可读指令被处理器执行时实现如上文所述的提高智能客服应答率的方法的步骤。
此外,为实现上述目的,本申请还提出一种提高智能客服应答率的装置,所述提高智能客服应答率的装置包括:分类模块、计算模块、聚类模块、获取模块和添加模块;
所述分类模块,设置为获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
所述计算模块,设置为计算各知识型问题之间的第一相似度;
所述聚类模块,设置为根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
所述获取模块,设置为将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
所述添加模块,设置为将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
附图说明
图1是本申请实施例方案涉及的硬件运行环境的提高智能客服应答率的设备结构示意图;
图2为本申请提高智能客服应答率的方法第一实施例的流程示意图;
图3为本申请提高智能客服应答率的方法第二实施例的流程示意图;
图4为本申请提高智能客服应答率的方法第三实施例的流程示意图;
图5为本申请提高智能客服应答率的装置第一实施例的结构框图。
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。
具体实施方式
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不设置为限定本申请。
参照图1,图1为本申请实施例方案涉及的硬件运行环境的提高智能客服应答率的设备结构示意图。
如图1所示,该提高智能客服应答率的设备可以包括:处理器1001,例如中央处理器(Central Processing Unit,CPU),通信总线1002、用户接口1003,网络接口1004,存储器1005。其中,通信总线1002设置为实现这些组件之间的连接通信。用户接口1003可以包括显示屏(Display),可选用户接口1003还可以包括标准的有线接口、无线接口,对于用户接口1003的有线接口在本申请中可为USB接口。网络接口1004可选的可以包括标准的有线接口、无线接口(如无线保真(WIreless-FIdelity,WI-FI)接口)。存储器1005可以是高速的随机存取存储器(Random Access Memory,RAM)存储器,也可以是稳定的存储器(Non-volatile Memory,NVM),例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储装置。
本领域技术人员可以理解,图1中示出的结构并不构成对提高智能客服应答率的设备的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件布置。
如图1所示,作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块、用户接口模块以及提高智能客服应答率的可读指令。
在图1所示的提高智能客服应答率的设备中,网络接口1004主要设置为连接后台服务器,与所述后台服务器进行数据通信;用户接口1003主要设置为连接智能客服设备;所述提高智能客服应答率的设备通过处理器1001调用存储器1005中存储的提高智能客服应答率的可读指令,并执行本申请实施例提供的提高智能客服应答率的方法。
基于上述硬件结构,提出本申请提高智能客服应答率的方法的实施例。
参照图2,图2为本申请提高智能客服应答率的方法第一实施例的流程示意图,提出本申请提高智能客服应答率的方法第一实施例。
在第一实施例中,所述提高智能客服应答率的方法包括以下步骤:
步骤S10:获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题。
应理解的是,本实施例的执行主体是提高智能客服应答率的设备,其中,所述提高智能客服应答率的设备可为个人电脑、服务器等电子设备。将未应答问题划分为两类,一类为知识型问题,一类为非知识型问题。知识型问题通常都能找到对应的标准答复,可将所述知识型问题补充到知识库中。非知识型问题,比如情绪、感情等感性问题,不具有标准回复,无需进行添加考虑,没有补充入知识库的必要。因此,我们经过分类处理后,保留未应答问题中的知识型问题,以对知识型问题做后续的添加处理。
需要说明的是,对所述未应答问题进行分类,可通过构建词表,所述词表可以看做是一个巨大的集合,知识型问题的训练集文本中所有词都保存在所述词表内,在构建过程中还可计算每个词的数量、索引和类型,所述类型包括语料和标签。在词向量空间中,这些知识型问题有近似的词向量,这些近似的词向量又对应相同的分类标签,在迭代训练过程中,这种相关性会不断传播,直至训练出准确的分类模型,则可通过所述分类模型对所述未应答问题进行分类,获得所述知识型问题。
步骤S20:计算各知识型问题之间的第一相似度。
在具体实现中,可对各知识型问题进行分词,计算词的TF-IDF值作为词特征,将各知识型问题表示为词向量,计算词向量间的余弦距离作为各知识型问题之间的第一相似度度量。TF-IDF实际上是:TF * IDF,TF词频(Term Frequency),IDF逆向文件频率(Inverse Document Frequency)。TF表示词条在文档d中出现的频率。IDF的主要思想是:如果包含词条t的文档越少,IDF越大,则说明词条t具有很好的类别区分能力。某一特定文档内的高词语频率,以及该词语在整个文档集合中的低文档频率,可以产生出高权重的TF-IDF。因此,TF-IDF倾向于过滤掉常见的词语,保留重要的词语。本实施例中,所述计算各知识型问题之间的第一相似度,具体包括:对各知识型问题进行分词处理,以获得所述知识型问题中所有的词语,计算所述词语的TF-IDF值;将各知识型问题表示为以词语和词语的TF-IDF值组成的词向量;计算各词向量之间的余弦距离,并将所述余弦距离作为各知识型问题之间的第一相似度。
步骤S30:根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇。
可理解的是,所述亲和传播聚类(Affinity Propagation,AP)算法,是根据数据点之间的相似度来进行聚类,可以是对称的,也可以是不对称的。所述亲和传播聚类算法不需要先确定聚类的数目,而是把所有的数据点都看成潜在意义上的聚类中心。根据各第一相似度度量可构建相似度矩阵,将各知识型问题作为节点,各知识型问题之间的第一相似度度量作为矩阵的值,然后经过AP算法进行聚类,获得多个聚类簇。
步骤S40:将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复。
需要说明的是,可从所述聚类簇中获取聚类中心,所述聚类中心与对应的聚类簇中的其他问题相似性最强,最具有该聚类簇的代表性,因此,选择聚类中心作为知识型问题的目标标准问题,聚类簇中的其他问题可作为所述目标标准问题对应的扩展问题。
应理解的是,知识型问题通常都能从现有的资料库中查找到相应的目标标准回复,可根据所述目标标准问题对现有的资料库进行查询,查询到与所述目标标准问题匹配的资料,将其作为目标标准回复。查询到与所述目标标准问题匹配的资料时,可进行人工干预,通过人工客服对匹配的资料进行整理,生成目标标准回复。
步骤S50:将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
可理解的是,将各聚类簇中最具代表性的聚类中心作为目标标准问题,将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,使得未应答问题中具有代表性的问题能够被添加至智能客服的知识库中。
在第一实施例中,获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题,从而剔除分类获得的非知识型问题,所述非知识型问题不具有标准回复的,无需进行添加考虑,从而提高后续添加的目标标准问题的质量;通过计算所述知识型问题之间的第一相似度,根据第一相似度通过亲和传播聚类算法获得多个聚类簇,将各聚类簇的聚类中心作为目标标准问题添加至智能客服的知识库中,所述聚类中心是各聚类簇中最具有代表性的问题,将其作为目标标准问题添加至智能客服中,能够有效提高智能客服的应答率,提升用户体验。
参照图3,图3为本申请提高智能客服应答率的方法第二实施例的流程示意图,基于上述图2所示的第一实施例,提出本申请提高智能客服应答率的方法的第二实施例。
在第二实施例中,所述步骤S10,包括:
步骤S101:获取智能客服的未应答问题,计算所述未应答问题的频率,将所述频率超过第一预设阈值的未应答问题作为待处理问题。
可理解的是,获取智能客服未应答问题,统计各未应答问题出现的次数,从而获得各未应答问题的频率,通常频率低的未应答问题属于长尾问题,问题质量通常偏低,不具有参考性,通常这类问题,我们不需要进行分析。所以,可预先设置所述第一预设阈值,超过所述第一预设阈值的未应答问题为客户比较关心的问题,经常问到,此类问题具有参考性,作为待处理问题,所述待处理问题若属于知识型问题,则找到对应的标准答复,可将其补充入知识库中,若下次客户再问及相关问题,则智能客服可从知识库中查找到对应的标准答复进行应答。
步骤S102:将所述待处理问题通过预设分类模型进行分类,获得知识型问题。
需要说明的是,所述预设分类模型包括快速文本(FastText)分类模型,所述FastText分类模型包含三部分:模型架构、分层分类器(Softmax)和汉语语言模型(N-gram)特征。分层分类器Softmax建立在哈弗曼编码的基础上,对标签进行编码,能够极大地缩小模型预测目标的数量,即极大缩小所述预设分类模型预测所述待处理问题中知识型问题的数量。FastText 加入了N-gram特征,来将局部词序考虑在内,从而实现对所述待处理问题更精确的分类。
在第二实施例中,所述步骤S40之后,还包括:
步骤S401:计算所述目标标准问题与所述智能客服的知识库中的原始标准问题之间的第二相似度。
应理解的是,可将知识库中各原始标准问题进行分词,计算词的TF-IDF值,将各原始标准问题表示为以词语和词语的TF-IDF值组成的词向量,计算各原始标准问题对应的词向量与所述目标标准问题对应的词向量之间的余弦距离,并将该余弦距离作为所述第二相似度。
步骤S402:获取所述第二相似度中的最大值,判断所述最大值是否超过第二预设阈值。
可理解的是,获取所述第二相似度中的最大值,该最大值对应的原始标准问题与所述目标标准问题最相近,判断所述最大值是否超过第二预设阈值,若超过所述第二预设阈值,说明所述目标标准问题与知识库中该最大值对应的原始标准问题相关性很高。
步骤S403:若所述最大值超过所述第二预设阈值,获取所述最大值对应的原始标准问题作为待补充问题。
需要说明的是,若所述最大值超过所述第二预设阈值,比如所述第二预设阈值为80%,若所述最大值为81%超过80%,则认为所述最大值对应的原始标准问题与所述目标标准问题相关性很高,甚至可能本质上是同一个问题,只是表述上存在差异,可建议将所述目标标准问题与该最大值对应的原始标准问题进行合并,将所述最大值对应的原始标准问题作为待补充问题,则将所述目标标准问题作为所述待补充问题的扩展问题补充到知识库中。
相应地,所述步骤S50,具体包括:
步骤S501:将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中所述待补充问题的扩展问题列表中。
在具体实现中,所述第二相似度中的最大值若超过所述第二预设阈值,说明所述最大值对应的原始标准问题与所述目标标准问题相关性很高,则可将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中所述待补充问题的扩展问题列表中,在用户问询到所述待补充问题或所述待补充问题的扩展问题列表中的某个问题时,智能客服可将所述待补充问题的扩展问题列表中的其他问题也进行展示,从而更全面的为用户提供想咨询问题的相关资料信息。
在第二实施例中,根据所述未应答问题的频率,将所述频率超过第一预设阈值的未应答问题作为待处理问题,从而将所述待处理问题进行分类获得知识型问题,将频率低的未应答问题进行了剔除,减少后续确定目标标准问题的工作量的同时提高知识库补充的目标标准问题的质量;计算所述目标标准问题与所述智能客服的知识库中的原始标准问题之间的第二相似度,若所述第二相似度中的最大值超过所述第二预设阈值,取所述最大值对应的原始标准问题作为待补充问题,将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中所述待补充问题的扩展问题列表中,从而更全面的为用户提供想咨询问题的相关资料信息。
参照图4,图4为本申请提高智能客服应答率的方法第三实施例的流程示意图,基于上述图3所示的第二实施例,提出本申请提高智能客服应答率的方法的第三实施例。
在第三实施例中,所述步骤S402之后,还包括:
步骤S404:若所述最大值未超过所述第二预设阈值,将各聚类簇中除了目标标准问题之外的其他知识型问题作为第一扩展问题。
应理解的是,若所述第二相似度中的最大值未超过所述第二预设阈值,说明所述最大值对应的原始标准问题与所述目标标准问题相关性不高,则可将所述目标标准问题以独立问题形式添加至所述知识库中。所述知识库中的各问题都是以原始标准问题及对应的扩展问题列表的形式存储,以便用户问询到某个原始标准问题或其扩展问题列表中的某个问题时,智能客服可将所述原始标准问题和对应的扩展问题列表中的其他问题也进行展示,从而更全面的为用户提供想咨询问题的相关资料信息。
可理解的是,所述目标标准问题可作为独立问题形式添加至所述知识库中,则所述目标标准问题对应的聚类簇中除了所述目标标准问题之外的其他知识型问题作为第一扩展问题,所述第一扩展问题可作为所述目标标准问题的扩展问题列表中的问题。
步骤S405:计算所述第一扩展问题与对应的目标标准问题之间的第三相似度。
需要说明的是,各聚类簇中的问题可能数量较多,无需将各聚类簇中的所有问题都作为所述目标标准问题的扩展问题一同添加至所述知识库中。则可计算所述第一扩展问题与对应的目标标准问题之间的第三相似度,将所述第三相似度的数值较大的第一扩展问题作为所述目标标准问题的扩展问题列表中的问题。
本实施例中,所述步骤S405,包括:计算所述第一扩展问题与对应的目标标准问题之间的统计特征、语义特征和主题特征;通过逻辑回归将所述统计特征、所述语义特征和所述主题特征进行聚合,获得所述第一扩展问题与对应的目标标准问题之间的第三相似度。其中,所述统计特征包括:词共现率、TF-IDF值、编辑距离和最长公共子串。可计算所述第一扩展问题与对应的目标标准问题之间的词共现率、TF-IDF值、编辑距离和最长公共子串作为所述统计特征。基于长短期记忆网络(Long Short-Term Memory,LSTM)构建所述第一扩展问题及所述目标标准问题的向量,计算所述向量之间的余弦相似度,将该余弦相似度作为所述语义特征。通过文档主题生成模型(Latent Dirichlet Allocation,LDA)对所述第一扩展问题和对应的目标标准问题进行对应的主题特征的生成。所述逻辑回归(Logistic Regression,LR)在线性回归的基础上,套用了一个逻辑函数,通过所述逻辑回归可将所述统计特征、所述语义特征和所述主题特征进行聚合,获得所述第一扩展问题与对应的目标标准问题之间的第三相似度。
步骤S406:按照所述第三相似度从大到小顺序在所述第一扩展问题中选择N个第一扩展问题作为所述目标标准问题的扩展问题列表,所述N为大于或等于1的整数。
在具体实现中,为了避免添加过多质量并不高的问题在所述目标标准问题的扩展问题列表中,通常可按照所述第三相似度从大到小顺序在所述第一扩展问题中选择N个第一扩展问题,所述N为大于或等于1的整数,也就是说,将与所述目标标准问题相似程度大的N个第一扩展问题作为所述目标标准问题的扩展问题列表。
在第三实施例中,所述步骤S50,具体包括:
步骤S502:将所述目标标准问题的扩展问题列表、所述目标标准问题及对应的目标标准回复以独立问题形式添加至所述智能客服的知识库中。
需要说明的是,所述第二相似度中的最大值未超过所述第二预设阈值,说明所述最大值对应的原始标准问题与所述目标标准问题相关性不高,则将所述目标标准问题的扩展问题列表、所述目标标准问题及对应的目标标准回复以独立问题形式添加至所述智能客服的知识库中,用户下次再问询所述目标标准问题相关的问题时,所述智能客服可从知识库中查找到所述目标标准问题及对应的扩展问题列表,从而实现应答用户。
在第三实施例中,若所述最大值未超过所述第二预设阈值,将各聚类簇中除了目标标准问题之外的其他知识型问题作为第一扩展问题,计算所述第一扩展问题与对应的目标标准问题之间的第三相似度,按照所述第三相似度从大到小顺序在所述第一扩展问题中选择N个第一扩展问题作为所述目标标准问题的扩展问题列表,将所述目标标准问题的扩展问题列表、所述目标标准问题及对应的目标标准回复以独立问题形式添加至所述智能客服的知识库中,使得用户下次再问询所述目标标准问题相关的问题时,所述智能客服可从知识库中查找到所述目标标准问题及对应的扩展问题列表,从而实现应答用户,提升用户体验。
此外,本申请实施例还提出一种存储介质,所述存储介质上存储有提高智能客服应答率的可读指令,所述提高智能客服应答率的可读指令被处理器执行时实现如上文所述的提高智能客服应答率的方法的步骤。所述存储介质可以为非易失性可读存储介质。
此外,参照图5,本申请实施例还提出一种提高智能客服应答率的装置,所述提高智能客服应答率的装置包括:分类模块10、计算模块20、聚类模块30、获取模块40和添加模块50;
所述分类模块10,设置为获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
所述计算模块20,设置为计算各知识型问题之间的第一相似度;
所述聚类模块30,设置为根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
所述获取模块40,设置为将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
所述添加模块50,设置为将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
本申请提高智能客服应答率的装置的其他实施例或具体实现方式可参照上述各方法实施例,此处不再赘述。
需要说明的是,在本文中,术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含,从而使得包括一系列要素的过程、方法、物品或者系统不仅包括那些要素,而且还包括没有明确列出的其他要素,或者是还包括为这种过程、方法、物品或者系统所固有的要素。在没有更多限制的情况下,由语句“包括一个……”限定的要素,并不排除在包括该要素的过程、方法、物品或者系统中还存在另外的相同要素。
上述本申请实施例序号仅仅为了描述,不代表实施例的优劣。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述 实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通 过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体 现出来,该计算机软件产品存储在一个存储介质(如只读存储器镜像(Read Only Memory image,ROM)/随机存取存储器(Random Access Memory,RAM)、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例,并非因此限制本申请的专利范围,凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换,或直接或间接运用在其他相关的技术领域,均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种提高智能客服应答率的方法,其中,所述提高智能客服应答率的方法包括以下步骤:
    获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
    计算各知识型问题之间的第一相似度;
    根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
    将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
    将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
  2. 如权利要求1所述的提高智能客服应答率的方法,其中,所述获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题,包括:
    获取智能客服的未应答问题,计算所述未应答问题的频率,将所述频率超过第一预设阈值的未应答问题作为待处理问题;
    将所述待处理问题通过预设分类模型进行分类,获得知识型问题。
  3. 如权利要求2所述的提高智能客服应答率的方法,其中,所述计算各知识型问题之间的第一相似度,包括:
    对各知识型问题进行分词处理,以获得所述知识型问题中所有的词语,计算所述词语的TF-IDF值;
    将各知识型问题表示为以词语和词语的TF-IDF值组成的词向量;
    计算各词向量之间的余弦距离,并将所述余弦距离作为各知识型问题之间的第一相似度。
  4. 如权利要求3所述的提高智能客服应答率的方法,其中,所述将所述聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复之后,所述提高智能客服应答率的方法还包括:
    计算所述目标标准问题与所述智能客服的知识库中的原始标准问题之间的第二相似度;
    获取所述第二相似度中的最大值,判断所述最大值是否超过第二预设阈值;
    若所述最大值超过所述第二预设阈值,获取所述最大值对应的原始标准问题作为待补充问题;
    所述将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,包括:
    将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中所述待补充问题的扩展问题列表中。
  5. 如权利要求4所述的提高智能客服应答率的方法,其中,所述获取所述第二相似度中的最大值,判断所述最大值是否超过第二预设阈值之后,所述提高智能客服应答率的方法还包括:
    若所述最大值未超过所述第二预设阈值,将各聚类簇中除了目标标准问题之外的其他知识型问题作为第一扩展问题;
    计算所述第一扩展问题与对应的目标标准问题之间的第三相似度;
    按照所述第三相似度从大到小顺序在所述第一扩展问题中选择N个第一扩展问题作为所述目标标准问题的扩展问题列表,所述N为大于或等于1的整数;
    所述将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,包括:
    将所述目标标准问题的扩展问题列表、所述目标标准问题及对应的目标标准回复以独立问题形式添加至所述智能客服的知识库中。
  6. 如权利要求5所述的提高智能客服应答率的方法,其中,所述计算所述第一扩展问题与对应的目标标准问题之间的第三相似度,包括:
    计算所述第一扩展问题与对应的目标标准问题之间的统计特征、语义特征和主题特征;
    通过逻辑回归将所述统计特征、所述语义特征和所述主题特征进行聚合,获得所述第一扩展问题与对应的目标标准问题之间的第三相似度。
  7. 如权利要求6所述的提高智能客服应答率的方法,其中,所述统计特征包括:词共现率、TF-IDF值、编辑距离和最长公共子串。
  8. 一种提高智能客服应答率的设备,其中,所述提高智能客服应答率的设备包括:存储器、处理器及存储在所述存储器上并可在所述处理器上运行的提高智能客服应答率的可读指令,所述提高智能客服应答率的可读指令配置为如下步骤:
    获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
    计算各知识型问题之间的第一相似度;
    根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
    将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
    将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
  9. 如权利要求8所述的提高智能客服应答率的设备,其中,所述提高智能客服应答率的可读指令配置为如下步骤:
    获取智能客服的未应答问题,计算所述未应答问题的频率,将所述频率超过第一预设阈值的未应答问题作为待处理问题;
    将所述待处理问题通过预设分类模型进行分类,获得知识型问题。
  10. 如权利要求9所述的提高智能客服应答率的设备,其中,所述提高智能客服应答率的可读指令配置为如下步骤:
    对各知识型问题进行分词处理,以获得所述知识型问题中所有的词语,计算所述词语的TF-IDF值;
    将各知识型问题表示为以词语和词语的TF-IDF值组成的词向量;
    计算各词向量之间的余弦距离,并将所述余弦距离作为各知识型问题之间的第一相似度。
  11. 如权利要求10所述的提高智能客服应答率的设备,其中,所述提高智能客服应答率的可读指令配置为如下步骤:
    计算所述目标标准问题与所述智能客服的知识库中的原始标准问题之间的第二相似度;
    获取所述第二相似度中的最大值,判断所述最大值是否超过第二预设阈值;
    若所述最大值超过所述第二预设阈值,获取所述最大值对应的原始标准问题作为待补充问题;
    所述将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,包括:
    将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中所述待补充问题的扩展问题列表中。
  12. 如权利要求11所述的提高智能客服应答率的设备,其中,所述提高智能客服应答率的可读指令配置为如下步骤:
    若所述最大值未超过所述第二预设阈值,将各聚类簇中除了目标标准问题之外的其他知识型问题作为第一扩展问题;
    计算所述第一扩展问题与对应的目标标准问题之间的第三相似度;
    按照所述第三相似度从大到小顺序在所述第一扩展问题中选择N个第一扩展问题作为所述目标标准问题的扩展问题列表,所述N为大于或等于1的整数;
    所述将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,包括:
    将所述目标标准问题的扩展问题列表、所述目标标准问题及对应的目标标准回复以独立问题形式添加至所述智能客服的知识库中。
  13. 如权利要求12所述的提高智能客服应答率的设备,其中,所述提高智能客服应答率的可读指令配置为如下步骤:
    计算所述第一扩展问题与对应的目标标准问题之间的统计特征、语义特征和主题特征;
    通过逻辑回归将所述统计特征、所述语义特征和所述主题特征进行聚合,获得所述第一扩展问题与对应的目标标准问题之间的第三相似度。
  14. 一种存储介质,其中,所述存储介质上存储有提高智能客服应答率的可读指令,所述提高智能客服应答率的可读指令被处理器执行时实现如下步骤:
    获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
    计算各知识型问题之间的第一相似度;
    根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
    将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
    将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
  15. 如权利要求14所述的存储介质,其中,所述提高智能客服应答率的可读指令被处理器执行时实现如下步骤:
    获取智能客服的未应答问题,计算所述未应答问题的频率,将所述频率超过第一预设阈值的未应答问题作为待处理问题;
    将所述待处理问题通过预设分类模型进行分类,获得知识型问题。
  16. 如权利要求15所述的存储介质,其中,所述提高智能客服应答率的可读指令被处理器执行时实现如下步骤:
    对各知识型问题进行分词处理,以获得所述知识型问题中所有的词语,计算所述词语的TF-IDF值;
    将各知识型问题表示为以词语和词语的TF-IDF值组成的词向量;
    计算各词向量之间的余弦距离,并将所述余弦距离作为各知识型问题之间的第一相似度。
  17. 如权利要求16所述的存储介质,其中,所述提高智能客服应答率的可读指令被处理器执行时实现如下步骤:
    计算所述目标标准问题与所述智能客服的知识库中的原始标准问题之间的第二相似度;
    获取所述第二相似度中的最大值,判断所述最大值是否超过第二预设阈值;
    若所述最大值超过所述第二预设阈值,获取所述最大值对应的原始标准问题作为待补充问题;
    所述将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,包括:
    将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中所述待补充问题的扩展问题列表中。
  18. 如权利要求17所述的存储介质,其中,所述提高智能客服应答率的可读指令被处理器执行时实现如下步骤:
    若所述最大值未超过所述第二预设阈值,将各聚类簇中除了目标标准问题之外的其他知识型问题作为第一扩展问题;
    计算所述第一扩展问题与对应的目标标准问题之间的第三相似度;
    按照所述第三相似度从大到小顺序在所述第一扩展问题中选择N个第一扩展问题作为所述目标标准问题的扩展问题列表,所述N为大于或等于1的整数;
    所述将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中,包括:
    将所述目标标准问题的扩展问题列表、所述目标标准问题及对应的目标标准回复以独立问题形式添加至所述智能客服的知识库中。
  19. 如权利要求18所述的存储介质,其中,所述提高智能客服应答率的可读指令被处理器执行时实现如下步骤:
    计算所述第一扩展问题与对应的目标标准问题之间的统计特征、语义特征和主题特征;
    通过逻辑回归将所述统计特征、所述语义特征和所述主题特征进行聚合,获得所述第一扩展问题与对应的目标标准问题之间的第三相似度。
  20. 一种提高智能客服应答率的装置,其中,所述提高智能客服应答率的装置包括:分类模块、计算模块、聚类模块、获取模块和添加模块;
    所述分类模块,设置为获取智能客服的未应答问题,对所述未应答问题进行分类,获得知识型问题;
    所述计算模块,设置为计算各知识型问题之间的第一相似度;
    所述聚类模块,设置为根据所述第一相似度通过亲和传播聚类算法将各知识型问题进行聚类,获得多个聚类簇;
    所述获取模块,设置为将各聚类簇的聚类中心作为目标标准问题,并获取与所述目标标准问题对应的目标标准回复;
    所述添加模块,设置为将所述目标标准问题及对应的目标标准回复添加至所述智能客服的知识库中。
PCT/CN2018/092989 2018-05-22 2018-06-27 提高智能客服应答率的方法、设备、存储介质及装置 WO2019223058A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810499843.9A CN108804567A (zh) 2018-05-22 2018-05-22 提高智能客服应答率的方法、设备、存储介质及装置
CN201810499843.9 2018-05-22

Publications (1)

Publication Number Publication Date
WO2019223058A1 true WO2019223058A1 (zh) 2019-11-28

Family

ID=64091417

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/092989 WO2019223058A1 (zh) 2018-05-22 2018-06-27 提高智能客服应答率的方法、设备、存储介质及装置

Country Status (2)

Country Link
CN (1) CN108804567A (zh)
WO (1) WO2019223058A1 (zh)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11735028B2 (en) 2018-06-12 2023-08-22 Intergraph Corporation Artificial intelligence applications for computer-aided dispatch systems
CN109688281A (zh) * 2018-12-03 2019-04-26 复旦大学 一种智能语音交互方法及系统
CN109783582B (zh) * 2018-12-04 2023-08-15 平安科技(深圳)有限公司 一种知识库对齐方法、装置、计算机设备及存储介质
CN109800879B (zh) * 2018-12-21 2022-02-01 科大讯飞股份有限公司 知识库构建方法和装置
CN111382235A (zh) * 2018-12-27 2020-07-07 上海智臻智能网络科技股份有限公司 一种问答知识库的优化方法及其装置
CN109885651B (zh) * 2019-01-16 2024-06-04 平安科技(深圳)有限公司 一种问题推送方法和装置
CN110555101A (zh) * 2019-09-09 2019-12-10 浙江诺诺网络科技有限公司 一种客服知识库更新方法、装置、设备及存储介质
CN110941708B (zh) * 2019-11-04 2023-02-28 智器云南京信息科技有限公司 智能问答库建立方法、智能问答方法及装置、计算机设备
CN110955766A (zh) * 2019-11-29 2020-04-03 支付宝(杭州)信息技术有限公司 一种自动扩充智能客服标准问题对的方法和系统
CN111159370A (zh) * 2019-12-20 2020-05-15 中国建设银行股份有限公司 一种短会话新问题生成方法、存储介质和人机交互装置
CN113392208A (zh) * 2020-03-12 2021-09-14 中国移动通信集团云南有限公司 It运维故障处理经验积累的方法、装置及存储介质
CN111488442A (zh) * 2020-04-09 2020-08-04 深圳追一科技有限公司 数据处理方法、装置、服务器及存储介质
CN111368060B (zh) * 2020-05-27 2021-01-26 支付宝(杭州)信息技术有限公司 对话机器人的自学习方法、装置、系统、电子设备及介质
CN111611781B (zh) * 2020-05-27 2023-08-18 北京妙医佳健康科技集团有限公司 数据标注方法、问答方法、装置及电子设备
CN113779050A (zh) * 2020-06-23 2021-12-10 北京沃东天骏信息技术有限公司 一种客服机器人知识库的管理方法和装置
CN111813910B (zh) * 2020-06-24 2024-05-31 平安科技(深圳)有限公司 客服问题的更新方法、系统、终端设备及计算机存储介质
CN111859985B (zh) * 2020-07-23 2023-09-12 上海华期信息技术有限责任公司 Ai客服模型测试方法、装置、电子设备及存储介质
CN112015899B (zh) * 2020-08-31 2023-08-11 中国银行股份有限公司 知识库的补充方法、装置、服务器和计算机存储介质
CN112148852A (zh) * 2020-09-10 2020-12-29 深圳康佳电子科技有限公司 一种智能客服方法、装置、存储介质及计算机设备
CN112115248B (zh) * 2020-09-27 2023-11-14 支付宝(杭州)信息技术有限公司 一种从对话语料中抽取对话策略结构的方法及系统
CN112541067A (zh) * 2020-12-15 2021-03-23 平安科技(深圳)有限公司 知识库问题挖掘方法、装置、电子设备及存储介质
CN112597292B (zh) * 2020-12-29 2024-04-26 招联消费金融股份有限公司 问题回复推荐方法、装置、计算机设备和存储介质
CN113139039A (zh) * 2021-05-06 2021-07-20 京东数字科技控股股份有限公司 对话数据处理方法、装置、设备以及存储介质
CN116743694A (zh) * 2022-03-02 2023-09-12 北京沃东天骏信息技术有限公司 一种电子邮件跟踪的方法和装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103810218A (zh) * 2012-11-14 2014-05-21 北京百度网讯科技有限公司 一种基于问题簇的自动问答方法和装置
CN104036015A (zh) * 2014-06-24 2014-09-10 北京奇虎科技有限公司 一种电子终端问题分类及解决方案提供方法、系统及装置
US20160179928A1 (en) * 2014-05-01 2016-06-23 International Business Machines Corporation Categorizing Users Based on Similarity of Posed Questions, Answers and Supporting Evidence
CN105955965A (zh) * 2016-06-21 2016-09-21 上海智臻智能网络科技股份有限公司 问句信息处理方法及装置
CN105975460A (zh) * 2016-05-30 2016-09-28 上海智臻智能网络科技股份有限公司 问句信息处理方法及装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102637192A (zh) * 2012-02-17 2012-08-15 清华大学 一种自然语言问答的方法
CN106547734B (zh) * 2016-10-21 2019-05-24 上海智臻智能网络科技股份有限公司 一种问句信息处理方法及装置
CN107656948B (zh) * 2016-11-14 2019-05-07 平安科技(深圳)有限公司 自动问答系统中的问题聚类处理方法及装置
CN107784105A (zh) * 2017-10-26 2018-03-09 平安科技(深圳)有限公司 基于海量问题的知识库构建方法、电子装置及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103810218A (zh) * 2012-11-14 2014-05-21 北京百度网讯科技有限公司 一种基于问题簇的自动问答方法和装置
US20160179928A1 (en) * 2014-05-01 2016-06-23 International Business Machines Corporation Categorizing Users Based on Similarity of Posed Questions, Answers and Supporting Evidence
CN104036015A (zh) * 2014-06-24 2014-09-10 北京奇虎科技有限公司 一种电子终端问题分类及解决方案提供方法、系统及装置
CN105975460A (zh) * 2016-05-30 2016-09-28 上海智臻智能网络科技股份有限公司 问句信息处理方法及装置
CN105955965A (zh) * 2016-06-21 2016-09-21 上海智臻智能网络科技股份有限公司 问句信息处理方法及装置

Also Published As

Publication number Publication date
CN108804567A (zh) 2018-11-13

Similar Documents

Publication Publication Date Title
WO2019223058A1 (zh) 提高智能客服应答率的方法、设备、存储介质及装置
WO2019242090A1 (zh) 智能客服应答方法、设备、存储介质及装置
WO2017150860A1 (en) Predicting text input based on user demographic information and context information
WO2017041484A1 (zh) 一种实时信息的推荐方法、装置和系统
WO2012134180A2 (ko) 문장에 내재한 감정 분석을 위한 감정 분류 방법 및 컨텍스트 정보를 이용한 다중 문장으로부터의 감정 분류 방법
WO2020138928A1 (en) Information processing method, apparatus, electrical device and readable storage medium
WO2018004236A1 (ko) 개인정보의 비식별화 방법 및 장치
WO2019019714A1 (zh) 一种微信客户行为反馈方法、设备及存储介质
WO2010024628A2 (ko) 확장 키워드 풀을 사용한 검색 방법 및 시스템
WO2010119996A1 (ko) 동영상 관련 광고를 제공하는 방법 및 그 장치
WO2019075944A1 (zh) 基于用户行为的信息屏蔽方法、装置、终端和存储介质
WO2020168606A1 (zh) 广告视频优化方法、装置、设备及计算机可读存储介质
WO2017213281A1 (ko) 빅데이터의 비식별화 처리 방법
WO2018164378A1 (en) Electronic apparatus for compressing language model, electronic apparatus for providing recommendation word and operation methods thereof
WO2018034426A1 (ko) 커널 rdr을 이용한 태깅 말뭉치 오류 자동수정방법
WO2020000621A1 (zh) 空调控制方法、空调控制装置、空调设备及存储介质
WO2019119624A1 (zh) 基于Excel表的计算方法、装置、设备及存储介质
WO2020113959A1 (zh) 医疗机构透析水平的考核方法、装置、设备及存储介质
WO2014000627A1 (zh) 网络信息推送方法及其系统、计算机存储介质
WO2018129978A1 (zh) 信息处理方法、装置、存储介质及计算机设备
WO2018233356A1 (zh) 文件扫描方法、系统、装置和计算机可读存储介质
WO2021029643A1 (en) System and method for modifying speech recognition result
WO2016129871A1 (ko) 영상처리장치, 영상처리장치의 제어방법 및 시스템
WO2019107674A1 (en) Computing apparatus and information input method of the computing apparatus
WO2021246812A1 (ko) 딥러닝 nlp 모델을 활용한 뉴스 긍정도 분석 솔루션 및 장치

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18920039

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 26.03.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18920039

Country of ref document: EP

Kind code of ref document: A1