CN111782793A - Intelligent customer service processing method, system and device - Google Patents

Abstract

The invention discloses an intelligent customer service processing method, system and equipment. The method includes: acquiring the content of multiple rounds of dialogue in the intelligent customer service system, performing model training according to the content of the multiple rounds of dialogue, and constructing a multi-category label classification model of the above context, wherein each label category corresponds to a reply item; The request content is sent to the multi-category label classification model for classification, and the corresponding reply items are output according to the classified labels. The solution of the present invention has higher flexibility and can handle more forms of user requests; compared with the customer service system based on the retrieval method, the use of a classification model to replace the retrieval method helps to improve the robustness of the overall system; it can solve the problem of the original retrieval method. The problem of the semantic gap that will be faced; it can reduce the influence of noise on the system's choice of reply items; the accuracy is higher.

Description

Intelligent customer service processing method, system and device

technical field

The invention relates to the field of computer technology, and in particular to a method, system and device for processing intelligent customer service.

Background technique

Spoken Dialogue System can be regarded as an interactive system that imitates human-human communication through natural language. ) technology has gradually matured and has been widely used in customer service systems. Since 2006, deep learning methods have been widely researched and applied, and many machine learning tasks have achieved breakthrough development. In 2013, word2vec, language models, and sequence-to-sequence models have appeared one after another. Massive amounts of data have been collected and applied to the system. training and design.

In the traffic card customer service scenario, the most important service is for business consulting such as card processing and activation, which belongs to the intelligent customer service system in the vertical field. The reply items are retrieved according to the user's input request. The traditional method is to use a series of manually predefined rules to perform semantic analysis and match the reply items. At present, the conventional practice is based on the retrieval-based model. The main idea of the retrieval-based dialogue system is to match the most suitable reply item among multiple reply candidates for the user's input request as the output. There are two matching methods. The first is a representation-based matching method. In the initial stage, text features are extracted for the input and reply candidates respectively, and then the similarity calculation is performed on the obtained text representations through the similarity function, and the one with the highest matching level is used as the output. The second method is an interaction-based matching method. Different from method 1, the similarity is calculated for the text representation in the final stage. The interaction-based matching method interacts with the text representation in the first stage of the model, and calculates through the similarity matrix. The model can obtain matching relationships of different levels and granularities, and finally returns the reply item with the highest matching level.

The problem existing in the prior art is that the method based on manual definition of rules is not flexible enough to well solve the diverse and changeable user requests; while the customer service system based on the retrieval method selects the most suitable reply item from the reply candidate set. As the output, the core of the method is to perform correlation matching between the user's input request information and the reply items (Relevance Matching). However, sometimes there is no direct relationship between the information and the reply, and there is a semantic gap. In addition, due to the information-reply matching, the quality of the two texts is very important. If there is irrelevant content in the text, such noise will easily affect the retrieval-based dialogue system when matching the reply items. Robustness lower.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide an intelligent customer service processing method, system and device, which are used to improve the above-mentioned defects in the prior art.

In order to achieve the above objects, the present invention adopts the following technical solutions.

In a first aspect, a method for processing intelligent customer service is provided, including: acquiring multiple rounds of dialogue content in an intelligent customer service system, performing model training according to the multiple rounds of dialogue content, and constructing a multi-category label classification model of the above context, where each label The class corresponds to a reply item; the request content input by the user is sent to the multi-class label classification model for classification, and the corresponding reply item is output according to the classified label.

In some possible implementations, the model training is performed according to the content of multiple rounds of dialogues to construct a multi-category label classification model of the above context, including: obtaining the content of the multiple rounds of dialogues from the intelligent customer service system; The content is associated with the above content, and the associated text data is labeled. The labeled categories include chat, business, and complaints. The business category is further labeled, and each category of business corresponds to a reply item; The text data is preprocessed, the preprocessing includes data cleaning and word segmentation processing, and the preprocessed text data is divided into a training set and a test set; text feature extraction is performed on the text data in the training set; the extracted text features are used to classify The model is trained to obtain the multi-class label classification model of the above context.

In some possible implementations, sending the request content input by the user into a multi-category label classification model for classification, and outputting corresponding reply items according to the classified labels, including: receiving the request content currently input by the user; The requested content is associated with the above content to generate input data; the generated input data is preprocessed, and the preprocessing includes data cleaning and word segmentation processing; text feature extraction is performed on the preprocessed input data; the extracted text The features are sent to the multi-class label classification model for classification, and the corresponding reply items are output according to the classified labels.

In a second aspect, an intelligent customer service system is provided, including: a classification module for acquiring multi-round dialogue content in the intelligent customer service system, performing model training according to the multi-round dialogue content, and constructing a multi-category label classification model of the above context, Each tag class corresponds to a reply item; the reply module is used to send the request content input by the user into the multi-category tag classification model for classification, and output the corresponding reply item according to the classified tags.

In some possible implementations, the classification module includes: an acquisition unit, used for acquiring multiple rounds of dialogue content from the intelligent customer service system, and storing it in a database; a labeling unit, used for associating the multiple rounds of dialogue content with the above content, Label the associated text data. The labeling categories include chat, business and complaints. The business category is further labelled. Each category of business corresponds to a reply item, and the labelled text data is stored in the database; preprocessing A unit for preprocessing the marked text data, the preprocessing includes data cleaning and word segmentation processing, and the preprocessed text data is divided into a training set and a test set; a text feature extraction unit is used for the training set. Extract the text features from the text data; the model building unit is used to train the classification model by using the extracted text features to obtain the multi-class label classification model of the above context.

In some possible implementations, the reply module includes: a receiving unit, configured to receive the request content currently input by the user; an association unit, configured to associate the request question currently input by the user with the above content to generate input data; a preprocessing unit for preprocessing the generated input data, the preprocessing includes data cleaning and word segmentation processing; a text feature extraction unit for performing text feature extraction on the preprocessed input data; an output unit for The extracted text features are sent to the multi-category label classification model for classification, and the corresponding reply items are output according to the classified labels.

In a third aspect, a computer device is provided, including a processor and a memory, the memory stores a program, the program includes computer-executable instructions, and when the computer device runs, the processor executes a program stored in the memory. The computer executes the instructions, so that the computer device executes the intelligent customer service processing method according to the first aspect.

In a fourth aspect, there is provided a computer-readable storage medium storing one or more programs, the one or more programs comprising computer-executable instructions that, when executed by a computer device, cause the computer device to perform the first The intelligent customer service processing method described in the aspect.

As can be seen from the above technical solutions, the embodiments of the present invention have the following advantages:

First, compared with the traditional manual rule method, the present invention is more flexible and can handle more forms of user requests;

Secondly, compared with the customer service system based on the retrieval method, the method of using the classification module to replace the retrieval method in the present invention helps to improve the robustness of the overall system;

Thirdly, because the text content of the reply item does not need to be used for matching, the present invention can solve the problem of semantic gap that the original retrieval system would face;

Thirdly, the present invention can simultaneously reduce the influence of noise on the system to select reply items;

Thirdly, compared with the existing intelligent customer service system, the present invention makes full use of the above dialogue content information, and responds to the current user request in a manner associated with the above, with higher accuracy.

Description of drawings

In order to explain the technical solutions of the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings that are used in the description of the embodiments and the prior art. Obviously, the drawings in the following description are only some implementations of the present invention. For example, for those of ordinary skill in the art, other drawings can also be obtained from these drawings without any creative effort.

1 is a schematic flowchart of a method for processing an intelligent customer service provided by an embodiment of the present invention;

2 is a structural diagram of an intelligent customer service system provided by an embodiment of the present invention;

3 is a flow chart of constructing a multi-label classification model in an embodiment of the present invention;

4 is a flow chart of text annotation in an embodiment of the present invention;

Fig. 5 is the flow chart of replying to the user part in the embodiment of the present invention;

FIG. 6 is a structural diagram of a computer device according to an embodiment of the present invention.

Detailed ways

In order to make those skilled in the art better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only Embodiments are part of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The terms "first", "second", "third" and the like in the description and claims of the present invention and the above drawings are used to distinguish different objects, rather than to describe a specific order. Furthermore, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes unlisted steps or units, or optionally also includes For other steps or units inherent to these processes, methods, products or devices.

The following detailed descriptions are given respectively through specific embodiments.

Embodiments of the present invention provide an intelligent customer service system and a corresponding processing method. The system is an end-to-end intelligent customer service system based on the classification method, which is mainly composed of two parts, one is the construction part of the multi-label classification model, and the other is the part of replying to the user. The construction part of the multi-label classification model is mainly to obtain the multi-round dialogue content in the intelligent customer service system, and perform text-related preprocessing, and then model the processed multi-round dialogue content as the multi-class label classification model of the above context. , each of which corresponds to a specific reply item. The reply to the user part is to process the request content input by the user, classify the request content of the user, and output the corresponding reply item.

Please refer to FIG. 1, an embodiment of the present invention provides a method for processing intelligent customer service, the method includes the following steps:

S1. Obtain multiple rounds of dialogue content in the intelligent customer service system, perform model training according to the multiple rounds of dialogue content, and construct a multi-category label classification model of the above context, where each label class corresponds to a reply item;

S2. Send the request content input by the user into a multi-category label classification model for classification, and output corresponding reply items according to the classified labels.

Further, in step S1, model training is performed according to the content of multiple rounds of dialogue, and a multi-category label classification model of the above context is constructed, which may include: obtaining the content of multiple rounds of dialogue from the intelligent customer service system; The content of the text is correlated, and the associated text data is marked. The marked categories include chat, business, and complaints. The business category is further marked by category, and each category of business corresponds to a reply item; Preprocessing, the preprocessing includes data cleaning and word segmentation, dividing the preprocessed text data into a training set and a test set; extracting text features from the text data in the training set; using the extracted text features to train the classification model , build a multi-class label classification model that obtains the above context.

Further, step S2 may include: receiving the request content currently input by the user; associating the request content currently input by the user with the above content to generate input data; preprocessing the generated input data, and the preprocessing includes data cleaning and word segmentation processing; extract text features from the preprocessed input data; send the extracted text features to a multi-category label classification model for classification, and output the corresponding reply items according to the classified labels.

Please refer to FIG. 2 , another embodiment of the present invention provides an intelligent customer service system, which includes:

The classification module 10 is used to obtain the content of multiple rounds of dialogue in the intelligent customer service system, perform model training according to the content of the multiple rounds of dialogue, and construct a multi-category label classification model of the above context, wherein each label class corresponds to a reply item.

The reply module 20 is configured to send the request content input by the user into the multi-class label classification model for classification, and output corresponding reply items according to the classified labels.

Further, the classification module 10 may include:

The acquisition unit is used to acquire the content of multiple rounds of conversations from the intelligent customer service system and store them in the database;

The labeling unit is used to associate the content of multiple rounds of conversations with the above content, and label the associated text data. The labeling categories include chat, business and complaints. The business category is further labelled, and each business category corresponds to A reply item that stores the marked text data in the database;

a preprocessing unit, used for preprocessing the marked text data, the preprocessing includes data cleaning and word segmentation processing, and the preprocessed text data is divided into a training set and a test set;

A text feature extraction unit, which is used to extract text features from the text data of the training set;

The model building unit is used for training the classification model by using the extracted text features, and constructing the multi-category label classification model obtained in the above context.

Further, the reply module 20 may include:

a receiving unit, configured to receive the request content currently input by the user;

an association unit, used to associate the request question currently input by the user with the above content to generate input data;

a preprocessing unit for preprocessing the generated input data, the preprocessing includes data cleaning and word segmentation;

A text feature extraction unit, which is used to extract text features from the preprocessed input data;

The output unit is used to send the extracted text features into the multi-category label classification model for classification, and output corresponding reply items according to the classified labels.

In the following, the construction part of the multi-label classification model and the part of replying to the user in the technical solutions of the embodiments of the present invention are respectively compared and further explained respectively.

(1) Construction of a multi-label classification model

Referring to Figure 3, the construction of a multi-label classification model includes the following steps:

1.1 Obtaining multi-round dialogue text data

A) Obtain online real conversation data from an intelligent customer service system, for example, a large traffic intelligent customer service system for the card scene of the ETC (Electronic Toll Collection, electronic non-stop toll collection system) card scene;

B) Store the acquired data in a database, for example: Mongo DB. MongoDB is a database based on distributed file storage.

1.2 Text information annotation

Referring to FIG. 4 , the process of text annotation is as follows.

A) First, mark the text data information of the multi-round dialogue content:

A.1) Import the multi-round dialogue content to be marked into the data marking system;

A.2) The data annotation personnel associate the annotation text of the current dialogue content with the above content;

A.3) Text annotation according to the content of the dialogue, the main categories can include small talk, complaints, business;

A.4) In the business category, category labeling can be further carried out, and each category corresponds to a reply item;

A.5) Submit annotation data.

B) Review the labeled data:

B.1) Reviewers review the marked data;

B.2) Whether to reject the labeled data, if so, the labeler will re-label the data and review it again;

B.3) Review and pass the data into the database for incremental training of the model.

C) All the labeled data are stored in the database, and the construction of the traffic card field dataset is completed.

1.3 Text Information Preprocessing

A) Text data cleaning, using existing tools, such as the re module in python, to remove special characters in text data in a regular way.

B) Text segmentation:

B.1) For the preprocessing of Chinese text, due to the structural relationship of Chinese, the feature granularity is that the word granularity is far superior to the word granularity, so it is necessary to perform word segmentation processing on Chinese;

B.2) The word segmentation method mainly includes the word segmentation method based on string matching, the word segmentation method based on statistics and the word segmentation method based on understanding, and can use existing tools, such as stammer word segmentation, to perform word segmentation processing on Chinese text;

C) Divide the preprocessed text data into a training set Tr_set and a test set Te_set in proportion.

1.4 Building a classification module

A) Text feature extraction:

A.1) Text feature engineering is mainly to extract feature representations that can reflect text features according to the database;

A.2) The input is the training set Tr_set of multi-round dialogue content processed by word segmentation, and existing engineering techniques such as TF-IDF can be used for text feature extraction;

A.3) TF-IDF (term frequency-inverse document frequency) is a commonly used weighting technique for information retrieval and data mining. TF is Term Frequency and IDF is Inverse Document Frequency.

Term Frequency (TF) is the number of times a term occurs divided by the total number of terms in the document. If the total number of words in a document is 100, and the word "activation" appears 3 times, then the word frequency of the word "activation" in the document is 0.03 (3/100). One way to calculate the document frequency (DF) is to measure how many documents have the word "active" and divide by the total number of documents contained in the document set. So, if the word "active" appears in 1,000 documents, and the total number of documents is 10,000,000, the document frequency is 0.0001 (1000/10,000,000). Finally, the TF-IDF score can be obtained by calculating the term frequency x (logarithm of the inverse of the document frequency). Taking the example above, the TF-IDF score for the word "activation" in this file set would be 0.12=(0.03xlog(1/0.0001)).

B) Classification model

B.1) The classification model used in this paper belongs to a supervised learning model, and the input is the multi-round dialogue feature representation q associated with the above and the corresponding business label l; here, the multi-round dialogue feature representation q, including the current round and the previous round Feature representation of several rounds of dialogue;

B.2) The classification algorithm adopts the existing supervised learning engineering technology such as support vector machine, and represents the training model according to the input text features;

B.3) Calculate the confidence level of each service label;

B.4) The classification model outputs business labels;

C) Export and save the multi-class label classification model, denoted as CM.

(2) Reply to the user

Referring to Figure 5, the Reply to User section includes the following steps:

2.1 User input request

The request content _qi input by the user is usually a question sentence containing a series of keywords, such as "abnormal card activation", "unable to connect to Bluetooth" and so on.

2.2 Associating the request entered by the user with the above content

Since the dialogue content in intelligent customer service often belongs to a multi-round question and answer context, the current round of user request content qi and the above content {q _in ,r _in ,...,q _{i -2} ,r _i-2 , q _i-1 , r _{i-1 }} are associated, where q is the request and r is the reply, and generating the input data qi ^' associated with the above can effectively help reply to the user request.

For the user request content q _i in the i-th round, not only q _i but also the dialogue data of n rounds before the i-th round must be input when classifying, namely {q _in ,r _in ,...,q _i-2 ,r _i-2 , q _i-1 , r _i-1 }. That is, the input data q _i ^' here includes not only q _i , but also {q _in ,r _in ,...,q _i-2 ,r _i-2 ,q _i-1 ,r _i-1 }. Among them, n is an empirical value, which can be determined according to specific needs, for example, it can be 3 or 4, etc., which is not limited.

2.3 Preprocessing of the request information _qi ^' associated with the above, please refer to step 1.3 above for details;

2.4 Feature extraction obtains _qi ^' , please refer to A) text feature extraction in the previous step 1.4 for details;

2.5 is sent into the classification model CM, and corresponds to the reply item _ri according to the business label 1 output by the model;

2.6 Output the reply item _ri to the user.

Referring to FIG. 6, an embodiment of the present invention further provides a computer device 60, including a processor 61 and a memory 62, the memory 62 stores a program, and the program includes computer execution instructions, when the computer device When the 60 is running, the processor 61 executes the computer-executed instructions stored in the memory, so that the computer device 60 executes the intelligent customer service processing method as described above.

An embodiment of the present invention also provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include computer-executable instructions that, when executed by a computer device, cause all The computer device executes the intelligent customer service processing method as described above.

To sum up, the embodiment of the present invention discloses an intelligent customer service processing method and system and related equipment. It can be seen from the above technical solutions that the embodiment of the present invention has the following advantages:

First, compared with the traditional manual rule method, the present invention is more flexible and can handle more forms of user requests;

Secondly, compared with the customer service system based on the retrieval method, the method of using the classification module to replace the retrieval method in the present invention helps to improve the robustness of the overall system;

Thirdly, because the text content of the reply item does not need to be used for matching, the present invention can solve the problem of semantic gap that the original retrieval system would face;

Thirdly, the present invention can simultaneously reduce the influence of noise on the system to select reply items;

Thirdly, compared with the existing intelligent customer service system, the present invention makes full use of the above dialogue content information, and responds to the current user request in a manner associated with the above, with higher accuracy.

In the above-mentioned embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.

The above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or modify some of the technical features. Equivalent replacements are made; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (8)

1. an intelligent customer service processing method, is characterized in that, comprises: Obtain the content of multiple rounds of conversations in the intelligent customer service system, perform model training according to the content of the multiple rounds of conversations, and build a multi-category label classification model in the context above, where each label class corresponds to a reply item; The request content input by the user is sent to the multi-category label classification model for classification, and the corresponding reply item is output according to the classified label. 2. The method according to claim 1, wherein the model training is performed according to the multi-round dialogue content, and the multi-class label classification model of the above context is constructed, comprising: Obtain multiple rounds of dialogue content from the intelligent customer service system; Associate the acquired multi-round conversation content with the above content, and annotate the associated text data. The annotated categories include small talk, business, and complaints. The business category is further category-marked, and each category of business corresponds to a reply item; Preprocessing the marked text data, the preprocessing includes data cleaning and word segmentation processing, and dividing the preprocessed text data into a training set and a test set; Extract text features from the text data of the training set; The classification model is trained with the extracted text features, and the multi-class label classification model of the above context is obtained. 3. The method according to claim 1, wherein the request content input by the user is sent into a multi-class label classification model for classification, and the corresponding reply item is output according to the classified label, comprising: Receive the request content currently input by the user; Associate the request content currently input by the user with the above content to generate input data; Preprocessing the generated input data, the preprocessing includes data cleaning and word segmentation; Extract text features from the preprocessed input data; The extracted text features are sent to the multi-category label classification model for classification, and the corresponding reply items are output according to the classified labels. 4. An intelligent customer service system, characterized in that, comprising: The classification module is used to obtain the content of multiple rounds of dialogues in the intelligent customer service system, perform model training according to the content of the multiple rounds of dialogues, and construct a multi-category label classification model in the above context, in which each label class corresponds to a reply item; The reply module is used to send the request content input by the user into the multi-category label classification model for classification, and output the corresponding reply item according to the classified label. 5. The system of claim 4, wherein the classification module comprises: The acquisition unit is used to acquire the content of multiple rounds of conversations from the intelligent customer service system and store them in the database; The labeling unit is used to associate the content of multiple rounds of dialogue with the above content, and label the associated text data. The labeling categories include chat, business and complaints. The business category is further labelled, and each business category corresponds to A reply item that stores the marked text data in the database; a preprocessing unit, used for preprocessing the marked text data, the preprocessing includes data cleaning and word segmentation processing, and the preprocessed text data is divided into a training set and a test set; A text feature extraction unit, which is used to extract text features from the text data of the training set; The model building unit is used to train the classification model by using the extracted text features to obtain the multi-class label classification model of the above context. 6. The system of claim 4, wherein the reply module comprises: a receiving unit, configured to receive the request content currently input by the user; an association unit, used to associate the request question currently input by the user with the above content to generate input data; a preprocessing unit for preprocessing the generated input data, the preprocessing includes data cleaning and word segmentation; A text feature extraction unit, which is used to extract text features from the preprocessed input data; The output unit is used to send the extracted text features into the multi-category label classification model for classification, and output corresponding reply items according to the classified labels. 7. A computer device comprising a processor and a memory, the memory having a program stored therein, the program comprising computer-executable instructions, when the computer device operates, the processor executes the computer stored in the memory The instruction is executed to cause the computer device to execute the intelligent customer service processing method according to any one of claims 1-3. 8. A computer-readable storage medium storing one or more programs comprising computer-executable instructions that, when executed by a computer device, cause the computer device to perform as claimed in the claims The intelligent customer service processing method described in any one of 1-3.

Images