Control method and system for intelligent customer service

Abstract

The invention provides a control method and system for intelligent customer service, and the method comprises the steps: calling a historical dialogue text corresponding to a current round of dialogue of a consulting object when a dialogue text inputted by the consulting object is received; if the historical dialogue text is successfully called, similar dialogue texts corresponding to the historical dialogue text are obtained from a preset dialogue analysis database, and the number of the similar dialogue texts is an integer larger than or equal to one; respectively carrying out semantic coding on the historical dialogue text and the similar dialogue text through a preset dialogue analyzer to obtain a corresponding similar semantic coding vector and a historical dialogue semantic coding vector; generating a dialogue reply text corresponding to the current round of dialogue of the consultation object; and outputting the dialogue reply text to the consultation object. The method does not need to depend on an artificially predefined reply template, and can improve the reply accuracy and the experience effect of the user on the intelligent customer service.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a control method and system for intelligent customer service. Background Art

[0002] With the rapid development of the Internet and mobile communication technology, the application of intelligent customer service is becoming more and more widespread in people's lives, and is used to respond to some user requests in order to better and more conveniently meet some of the user's life needs. For example, it is commonly used in auxiliary power resource maintenance systems to simulate the behavior of staff who maintain power supply-related businesses to query power resource consumption or data resource payment, etc. It can determine the power resources consumed by the user's corresponding target area through dialogue and give corresponding data resource payment suggestions, etc., which can not only help users avoid the time and money consumption caused by waiting in line at the power supply service place, but also pay attention to the use of power resources in the user's corresponding target area in a timely manner.

[0003] The current auxiliary power resource maintenance system is usually implemented by using the technical means of language template generation of conventional power business, wherein the language template generation of conventional power business is generally achieved by pre-defining language templates of power business with identifiers, and generating corresponding keywords based on the dialogue identifier management mode to get replies. The dialogue identifier management mode is usually used to decide whether the next round will be a user query or a query result.

[0004] However, most of the language template generation methods for conventional power business need to rely on manually predefined language templates for conventional power business, and can only provide feedback on very limited user-related power business requests, and the language diversity of the responses is low, which leads to reduced accuracy of the responses and easily leads to poor user experience. Summary of the invention

[0005] The purpose of the present invention is to propose a control method and system for intelligent customer service to solve the technical problem of how to improve the accuracy of replies and the user experience of intelligent customer service.

[0006] On the one hand, a control method for intelligent customer service is provided, comprising:

[0007] When receiving the dialogue text input by the consulting object, retrieving the historical dialogue text corresponding to the current round of dialogue of the consulting object;

[0008] If the historical conversation text is retrieved successfully, similar conversation texts corresponding to the historical conversation text are obtained from a preset conversation analysis database, wherein the number of similar conversation texts is an integer greater than or equal to one;

[0009] Semantically encoding the historical conversation text and the similar conversation text respectively through a preset conversation analyzer to obtain corresponding similar semantic encoding vectors and historical conversation semantic encoding vectors;

[0010] Generate a dialogue reply text corresponding to the current round of dialogue of the consulting object according to the similar semantic coding vector and the historical dialogue semantic coding vector;

[0011] The dialogue reply text is output to the consultation object.

[0012] Preferably, the method further comprises: if the historical conversation text is not retrieved, obtaining a similar conversation template corresponding to the current round of conversation from a conversation analysis database;

[0013] A corresponding dialogue reply template is determined according to the similar dialogue template, and the dialogue reply template is output to the consultation object as a dialogue reply text corresponding to the current round of dialogue of the consultation object.

[0014] Preferably, the obtaining of similar conversation texts corresponding to the historical conversation texts includes traversing business keywords of the historical conversation texts and matching conversation text sets corresponding to the business keywords from the conversation analysis database;

[0015] The similar dialogue text is selected from the dialogue text collection.

[0016] Preferably, matching the conversation text set corresponding to the business keyword from the conversation analysis database includes traversing the historical conversation text based on a preset semantic search engine to obtain at least one business keyword;

[0017] According to the business keyword, the business dialogue text corresponding to the business keyword is searched from the dialogue database to obtain a dialogue text set corresponding to the scene.

[0018] Preferably, the conversation analyzer is specifically used to splice the similar conversation texts respectively to obtain a first long text;

[0019] Generate a corresponding similar semantic encoding vector according to the first long text.

[0020] Preferably, the conversation analyzer is further used to concatenate sentences in the historical conversation text to obtain a second long text;

[0021] Generate a corresponding historical conversation semantic encoding vector based on the second long text.

[0022] Preferably, generating the dialogue reply text corresponding to the current round of dialogue of the consulting object includes inputting the similar semantic encoding vector and the historical dialogue semantic encoding vector into a preset dialogue predictor, and generating at least one dialogue prediction sequence and sequence score through the dialogue predictor;

[0023] From all the dialogue prediction sequences, the dialogue prediction sequence corresponding to the maximum sequence value is selected as the dialogue reply text corresponding to the current round of dialogue of the consulting object.

[0024] Preferably, the method further comprises, when the historical conversation text is not retrieved, traversing the conversation keywords of the current round of conversation;

[0025] From the conversation analysis database, similar conversation templates corresponding to the conversation keywords are matched.

[0026] Preferably, matching the similar dialogue template corresponding to the dialogue keyword includes calculating the similarity between the dialogue keyword and each dialogue template;

[0027] The dialogue template corresponding to the maximum similarity is used as the similar dialogue template.

[0028] On the other hand, a control system for intelligent customer service is also provided to implement the control method for intelligent customer service, including:

[0029] A text acquisition module, for retrieving the historical conversation text corresponding to the current round of conversation of the consulting object when receiving the conversation text input by the consulting object;

[0030] If the historical conversation text is retrieved successfully, similar conversation texts corresponding to the historical conversation text are obtained from a preset conversation analysis database, wherein the number of similar conversation texts is an integer greater than or equal to one;

[0031] A semantic coding module, used to perform semantic coding on the historical conversation text and the similar conversation text respectively through a preset conversation analyzer to obtain corresponding similar semantic coding vectors and historical conversation semantic coding vectors;

[0032] A dialogue generation module, used to generate a dialogue reply text corresponding to the current round of dialogue of the consulting object according to the similar semantic coding vector and the historical dialogue semantic coding vector;

[0033] The dialogue reply text is output to the consultation object.

[0034] In summary, the implementation of the embodiments of the present invention has the following beneficial effects:

[0035] The control method and system for intelligent customer service provided by the present invention automatically retrieve a number of similar conversation samples from a large-scale conversation analysis database, so that the intelligent customer service can quickly obtain similar semantic coding vectors and historical conversation semantic coding vectors that fully express the semantic features of the conversation text through a conversation analyzer, and then can use the similar semantic coding vectors to more accurately assist the historical conversation coding vectors in generating conversation reply text corresponding to the current round of conversation, so that the intelligent customer service does not need to rely on manually predefined reply templates, and can improve the accuracy of the reply and the user experience of the intelligent customer service. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying creative labor, other drawings obtained based on these drawings still belong to the scope of the present invention.

[0037] Figure 1 The present invention is a schematic diagram of the main process of a control method for intelligent customer service in an embodiment of the present invention.

[0038] Figure 2 A schematic diagram of an exemplary system architecture in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings.

[0040] like Figure 1 FIG. 1 is a schematic diagram of an embodiment of a control method for intelligent customer service provided by the present invention. In this embodiment, the method includes the following steps:

[0041] Step S1, when receiving the conversation text input by the consulting object, retrieve the historical conversation text corresponding to the current round of conversation of the consulting object; wherein, the historical conversation text can specifically be expressed in the conversation history of other rounds (i.e. K-1 times) before the current round (assuming K times), such as text or images, etc., and no specific limitation is made here.

[0042] Specifically, assuming that the current round of dialogue on the power supply service page of the consulting object reaches the Kth round, the historical dialogue text of the previous K-1 rounds can be obtained, that is, the historical dialogue text corresponding to the current round of dialogue on the power supply service page of the consulting object. For example, assuming that the power supply service page of the current consulting object inquires the control system of the intelligent customer service about the power resource consumption of family A in the current month, and the current dialogue reaches the Kth round, assuming that K is 2, then the historical dialogue text of the previous K-1 rounds is obtained. Since K-1 is 1, the historical dialogue text of the previous K-1 rounds (i.e., the first round) is obtained, and all dialogues between the consulting object and the control system of the intelligent customer service on the power supply service page of the current consulting object are obtained, such as dialogue 1 of the control system of the intelligent customer service, "Hello, I am a service assistant of the control system of the intelligent customer service, and I need to understand your business needs first in order to facilitate the query", and dialogue 2 "the power resource consumption of family A in the current month" replied by the query object after dialogue 1, etc., as the historical dialogue text corresponding to the current round of dialogue.

[0043] It should be noted that if Figure 2 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0044] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0045] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV) players, laptop computers, desktop computers, etc.

[0046] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .

[0047] It should be noted that the control method for intelligent customer service based on conversation analysis provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the control system of the intelligent customer service is generally set in the server / terminal device.

[0048] It should be understood that Figure 2 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0049] In this embodiment, the electronic device (eg Figure 2 The server / terminal device shown in the figure can receive the conversation text through a wired connection or a wireless connection. It should be noted that the above-mentioned wireless connection method may include but is not limited to 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or to be developed in the future.

[0050] In one embodiment, in an actual application scenario, for example, when a consulting object participating in a conversation has a conversation with an intelligent customer service of a control system (for example, an intelligent customer service of an intelligent power supply APP), when entering the current round, the intelligent customer service of the control system needs to automatically reply in the current round of conversation with a reply dialogue suitable for the current conversation content (for example, a query dialogue on the electricity resource consumption of household A this month, or a query dialogue on the electricity resource payment of household B last month, etc.), that is, the conversation reply text. The historical conversation text corresponding to the current round of conversation of the power supply business service page of the consulting object can be obtained first, so that the conversation reply text corresponding to the current round of conversation can be more accurately predicted based on the historical conversation text corresponding to the current round of conversation of the power supply business service page of the consulting object.

[0051] Step S2, if the historical conversation text is retrieved successfully, then similar conversation texts corresponding to the historical conversation text are obtained from the preset conversation analysis database, wherein the number of similar conversation texts is an integer greater than or equal to one; it is understandable that if there is a historical conversation text, N similar conversation texts corresponding to the historical conversation text are obtained from the conversation analysis database, wherein N is an integer greater than or equal to 1. In this embodiment, after the historical conversation text is obtained, it can be searched in the conversation analysis database based on the historical conversation text, specifically by calculating the similarity between the historical conversation text and each text in the database through a similarity calculation formula, such as a cosine similarity algorithm or an Euclidean distance formula, and then filtering texts with a similarity greater than or equal to a similarity threshold as N similar conversation texts, or other retrieval methods can also be used, which are not specifically limited here. For example, it is assumed that a similar conversation text similar to the historical conversation text "Household A's Electricity Resource Consumption in January" is "A District's Electricity Resource Consumption in January.

[0052] In one embodiment, the acquisition of similar conversation texts corresponding to the historical conversation texts includes traversing the business keywords of the historical conversation texts, and matching the conversation text set corresponding to the business keywords from the conversation analysis database; and filtering out the similar conversation texts from the conversation text set. The matching of the conversation text set corresponding to the business keyword from the conversation analysis database includes traversing the historical conversation texts based on a preset semantic search engine to obtain at least one business keyword; searching the conversation database for business conversation texts corresponding to the business keyword according to the business keyword to obtain a conversation text set corresponding to the scene. It is understandable that the business keywords of the historical conversation texts are traversed, and the conversation text set corresponding to the business keywords is matched from the conversation analysis database; and N similar conversation texts are filtered out from the conversation text set. Based on the semantic search engine, the intelligent customer service traverses the historical conversation texts to obtain at least one business keyword; based on the at least one business keyword, the business conversation text corresponding to the at least one business keyword is searched from the conversation database to obtain a scene conversation text set.

[0053] One embodiment also includes, if the historical conversation text is not retrieved, obtaining a similar conversation template corresponding to the current round of conversation from a conversation analysis database; determining a corresponding conversation reply template based on the similar conversation template, and outputting the conversation reply template to the consulting object as the conversation reply text corresponding to the current round of conversation of the consulting object.

[0054] Specifically, when the historical conversation text is not retrieved, the conversation keywords of the current round of conversation are traversed; and the similar conversation templates corresponding to the conversation keywords are matched from the conversation analysis database. The matching of the similar conversation templates corresponding to the conversation keywords includes calculating the similarity between the conversation keywords and each conversation template; and taking the conversation template corresponding to the maximum similarity as the similar conversation template. Assuming that the historical conversation text is not obtained, a search can be performed in the conversation analysis database based on the historical conversation text. Specifically, each word in the current round of conversation is traversed first, and then the conversation keywords of the current round of conversation are obtained based on the vocabulary. Further, similar conversation templates corresponding to the conversation keywords can be matched from the conversation analysis database. Specifically, the similarity between the conversation keywords and each conversation template in the database can be calculated by a similarity calculation formula, such as a cosine similarity algorithm or an Euclidean distance formula, and then the templates with a similarity greater than or equal to a similarity threshold are screened as similar conversation templates. Alternatively, other search methods can be used, which are not specifically limited here.

[0055] Step S3, semantically encode the historical conversation text and the similar conversation text respectively through a preset conversation analyzer to obtain corresponding similar semantic encoding vectors and historical conversation semantic encoding vectors; it can be understood that based on the conversation analyzer, semantic encoding is performed on N similar conversation texts respectively to obtain similar semantic encoding vectors corresponding to the N similar conversation texts, and the historical conversation text is semantically encoded to obtain the historical conversation semantic encoding vector corresponding to the historical conversation text; after obtaining the N similar conversation texts corresponding to the historical conversation text, the N similar conversation texts can be input into a semantic encoding module, and the semantic encoding module uses the conversation analyzer to encode the N similar conversation texts to obtain similar semantic encoding vectors corresponding to the N similar conversation texts (for example, similar semantic encoding vectors are obtained through a GRU encoder).

[0056] In one embodiment, the conversation analyzer is specifically used to splice the similar conversation texts respectively to obtain a first long text; and generate a corresponding similar semantic encoding vector according to the first long text. The conversation analyzer is also used to splice the sentences in the historical conversation text to obtain a second long text; and generate a corresponding historical conversation semantic encoding vector according to the second long text. After obtaining the historical conversation text, the historical conversation text can be encoded by a semantic encoding module, specifically by first splicing each sentence in the historical conversation text into a long text, and then inputting the long text into an encoder for encoding to obtain the historical conversation encoding vector corresponding to the historical conversation text. Splice N similar conversation texts respectively to obtain a first long text; input the first long text into the conversation analyzer to generate a similar semantic encoding vector; splice the sentences in the historical conversation text to obtain a second long text; input the second long text into the conversation analyzer to generate a historical conversation semantic encoding vector.

[0057] Step S4, based on the similar semantic coding vector and the historical conversation semantic coding vector, generate a conversation reply text corresponding to the current round of conversation of the consulting object; and output the conversation reply text to the consulting object. It can be understood that based on the similar semantic coding vector and the historical conversation semantic coding vector, the intelligent customer service automatically generates a conversation reply text corresponding to the current round of conversation on the power supply business service page of the consulting object.

[0058] In one embodiment, the generation of the dialogue reply text corresponding to the current round of dialogue of the consulting object includes: inputting the similar semantic coding vector and the historical dialogue semantic coding vector into a preset dialogue predictor, generating at least one dialogue prediction sequence and sequence score through the dialogue predictor; and selecting the dialogue prediction sequence corresponding to the maximum sequence value from all dialogue prediction sequences as the dialogue reply text corresponding to the current round of dialogue of the consulting object. It can be understood that after the similar semantic coding vector and the historical dialogue semantic coding vector, the similar semantic coding vector and the historical dialogue semantic coding vector can be input into the dialogue predictor together to predict the dialogue reply, specifically, the dialogue predictor can be used to generate dialogue prediction sequences and sequence scores corresponding to several time steps, and then, the sequence scores of several dialogue prediction sequences can be compared, and the dialogue prediction sequence corresponding to the maximum sequence value can be selected from them to serve as the dialogue reply text corresponding to the current round of dialogue of the power supply business service page of the consulting object.

[0059] In this embodiment, a dialogue reply template corresponding to a similar dialogue template is obtained, and the intelligent customer service automatically uses the dialogue reply template as the dialogue reply text corresponding to the current round of dialogue on the power supply business service page of the consulting object. After obtaining the similar dialogue template corresponding to the current round of dialogue, the dialogue reply template corresponding to the similar dialogue template can be matched in the dialogue analysis database, that is, the reply template of the experience summary, so that the intelligent customer service automatically uses the dialogue reply template as the dialogue reply text corresponding to the current round of dialogue on the power supply business service page of the consulting object.

[0060] An embodiment of the present invention further provides a control system for intelligent customer service, which is used to implement the control method for intelligent customer service, including:

[0061] A text acquisition module, for retrieving the historical conversation text corresponding to the current round of conversation of the consulting object when receiving the conversation text input by the consulting object;

[0062] If the historical conversation text is retrieved successfully, similar conversation texts corresponding to the historical conversation text are obtained from a preset conversation analysis database, wherein the number of similar conversation texts is an integer greater than or equal to one;

[0063] A semantic coding module, used to perform semantic coding on the historical conversation text and the similar conversation text respectively through a preset conversation analyzer to obtain corresponding similar semantic coding vectors and historical conversation semantic coding vectors;

[0064] A dialogue generation module, used to generate a dialogue reply text corresponding to the current round of dialogue of the consulting object according to the similar semantic coding vector and the historical dialogue semantic coding vector;

[0065] The dialogue reply text is output to the consultation object.

[0066] It should be noted that the system described in the above embodiment corresponds to the method described in the above embodiment. Therefore, the undetailed parts of the system described in the above embodiment can be obtained by referring to the contents of the method described in the above embodiment, and will not be repeated here.

[0067] In summary, the implementation of the embodiments of the present invention has the following beneficial effects:

[0068] The control method and system for intelligent customer service provided by the present invention automatically retrieve a number of similar conversation samples from a large-scale conversation analysis database, so that the intelligent customer service can quickly obtain similar semantic coding vectors and historical conversation semantic coding vectors that fully express the semantic features of the conversation text through a conversation analyzer, and then can use the similar semantic coding vectors to more accurately assist the historical conversation coding vectors in generating conversation reply text corresponding to the current round of conversation, so that the intelligent customer service does not need to rely on manually predefined reply templates, and can improve the accuracy of the reply and the user experience of the intelligent customer service.

[0069] The above disclosure is only the preferred embodiment of the present invention, which certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A control method for intelligent customer service, characterized in that: include: When receiving the dialogue text input by the consulting object, retrieving the historical dialogue text corresponding to the current round of dialogue of the consulting object; If the historical conversation text is retrieved successfully, similar conversation texts corresponding to the historical conversation text are obtained from a preset conversation analysis database, wherein the number of similar conversation texts is an integer greater than or equal to one; Semantically encoding the historical conversation text and the similar conversation text respectively through a preset conversation analyzer to obtain corresponding similar semantic encoding vectors and historical conversation semantic encoding vectors; Generate a dialogue reply text corresponding to the current round of dialogue of the consulting object according to the similar semantic coding vector and the historical dialogue semantic coding vector; The dialogue reply text is output to the consultation object.

2. The method according to claim 1, characterized in that Also includes, If the historical conversation text is not retrieved, a similar conversation template corresponding to the current round of conversation is obtained from the conversation analysis database; A corresponding dialogue reply template is determined according to the similar dialogue template, and the dialogue reply template is output to the consultation object as a dialogue reply text corresponding to the current round of dialogue of the consultation object.

3. The method according to claim 2, characterized in that The obtaining of similar conversation texts corresponding to the historical conversation texts includes traversing business keywords of the historical conversation texts and matching conversation text sets corresponding to the business keywords from the conversation analysis database; The similar dialogue text is selected from the dialogue text collection.

4. The method according to claim 3, characterized in that The matching of the conversation text set corresponding to the business keyword from the conversation analysis database includes traversing the historical conversation text based on a preset semantic search engine to obtain at least one business keyword; According to the business keyword, the business dialogue text corresponding to the business keyword is searched from the dialogue database to obtain a dialogue text set corresponding to the scene.

5. The method according to claim 4, characterized in that The conversation analyzer is specifically used to splice the similar conversation texts respectively to obtain a first long text; Generate a corresponding similar semantic encoding vector according to the first long text.

6. The method according to claim 5, characterized in that The conversation analyzer is further used to concatenate the sentences in the historical conversation text to obtain a second long text; Generate a corresponding historical conversation semantic encoding vector based on the second long text.

7. The method according to claim 6, characterized in that Generating the dialogue reply text corresponding to the current round of dialogue of the consulting object includes inputting the similar semantic encoding vector and the historical dialogue semantic encoding vector into a preset dialogue predictor, and generating at least one dialogue prediction sequence and a sequence score through the dialogue predictor; From all the dialogue prediction sequences, the dialogue prediction sequence corresponding to the maximum sequence value is selected as the dialogue reply text corresponding to the current round of dialogue of the consulting object.

8. The method according to claim 2, characterized in that The method further includes, when the historical conversation text is not retrieved, traversing the conversation keywords of the current round of conversation; From the conversation analysis database, similar conversation templates corresponding to the conversation keywords are matched.

9. The method according to claim 8, characterized in that The matching of the similar dialogue templates corresponding to the dialogue keyword includes calculating the similarity between the dialogue keyword and each dialogue template; The dialogue template corresponding to the maximum similarity is used as the similar dialogue template.

10. A control system for intelligent customer service, used to implement the method according to any one of claims 1 to 9, characterized in that: include: A text acquisition module, for retrieving the historical conversation text corresponding to the current round of conversation of the consulting object when receiving the conversation text input by the consulting object; If the historical conversation text is retrieved successfully, similar conversation texts corresponding to the historical conversation text are obtained from a preset conversation analysis database, wherein the number of similar conversation texts is an integer greater than or equal to one; A semantic coding module, used to perform semantic coding on the historical conversation text and the similar conversation text respectively through a preset conversation analyzer to obtain corresponding similar semantic coding vectors and historical conversation semantic coding vectors; A dialogue generation module, used to generate a dialogue reply text corresponding to the current round of dialogue of the consulting object according to the similar semantic coding vector and the historical dialogue semantic coding vector; The dialogue reply text is output to the consultation object.

Images