CN115640386A

CN115640386A - Method and apparatus for conducting dialogs based on recommended dialogs

Info

Publication number: CN115640386A
Application number: CN202211047003.1A
Authority: CN
Inventors: 凌悦; 付宇
Original assignee: Shengdoushi Shanghai Technology Development Co Ltd
Current assignee: Shengdoushi Shanghai Science and Technology Development Co Ltd
Priority date: 2022-08-30
Filing date: 2022-08-30
Publication date: 2023-01-24

Abstract

The application provides a method and a device for conversation based on recommended speech. The method comprises the following steps: selecting a first recommended dialog template from at least one recommended dialog template corresponding to the current dialog topic based on the real-time dialog data and prior information of dialog participants for the current dialog topic in one or more dialog topics involved in the real-time dialog, wherein the dialog topic is an intention or purpose of the real-time dialog; generating conversation content based on the first recommended conversation template; obtaining a reply of a conversation participant to the generated conversation content, and updating the real-time conversation data based on the reply to generate next conversation content, wherein a conversation topic and at least one recommended conversation template corresponding to the conversation topic are determined based on historical conversation data of historical conversations.

Description

Method and apparatus for conducting dialogs based on recommended dialogs

Technical Field

The present application relates to intelligent information interaction, and more particularly, to a method, apparatus, and computer-readable medium for intelligent dialogue based on recommended language techniques.

Background

With the development of speech recognition and semantic matching technologies, there are more and more applications of using a conversation robot to complete communication based on a conversational recommendation algorithm when a conversation is performed with a human user. Common intelligent conversation schemes often adopt fixed recommendation dialogs, and cannot obtain conversation requirements of human users in real time and respond in time, so that the effect of automatic conversation is not ideal.

Accordingly, there is a need for improvements to existing intelligent dialog schemes.

Disclosure of Invention

The application provides a method and equipment for selecting recommended dialogs based on prior information in historical dialog data and real-time dialog data.

According to an aspect of the present application, a method for conducting a conversation based on recommended speech technology is provided, including:

selecting, for a current conversation topic of one or more conversation topics involved in a real-time conversation, a first recommended conversation template from at least one recommended conversation template corresponding to the current conversation topic based on real-time conversation data and prior information of conversation participants, wherein the conversation topic is used to indicate an intention or purpose of the real-time conversation;

generating conversation content based on the first recommended conversation template;

obtaining a reply from a conversation participant to the generated conversation content, and updating the real-time conversation data based on the reply to generate a next conversation content,

wherein the conversation topic and the at least one recommended conversation template corresponding to the conversation topic are determined based on historical conversation data of historical conversations.

According to another aspect of the present application, there is provided an apparatus for conducting a conversation based on recommended speech, comprising: the interaction unit is configured to acquire real-time conversation data and output conversation contents to conversation participants; and a recommendation unit configured to select, for a current conversation topic of the one or more conversation topics involved in the real-time conversation, a first recommended conversation template from at least one recommended conversation template corresponding to the current conversation topic based on the real-time conversation data and prior information of conversation participants, wherein the conversation topic is used for indicating an intention or purpose of the real-time conversation; generating conversation content based on the first recommended conversation template; and obtaining a reply from a conversation participant to the generated conversation content, and updating the real-time conversation data based on the reply to generate a next conversation content,

According to yet another aspect of the application, a computer-readable storage medium is proposed, on which a computer program is stored, the computer program comprising executable instructions that, when executed by a processor, carry out the method as described above.

According to yet another aspect of the present application, an electronic device is provided, comprising a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute the executable instructions to implement the method as described above.

The proposal for carrying out the conversation based on the recommended dialogues, which is provided by the application, can combine the prior information associated with the conversation participants and the conversation scenes and the real-time conversation data of the current conversation to timely and accurately acquire the conversation requirements and feedback information of the human conversation participants, select a more appropriate recommended dialogues template to ask questions of the conversation participants or reply questions of the conversation participants, quickly acquire the information required by the conversation tasks from the conversation, and continuously keep good communication effect with the human conversation participants before the conversation tasks are completed. On the basis of immediately and accurately acquiring the requirements of the interlocutors and responding, the scheme of the application can also return real-time dialogue data periodically according to the dialogue effect of the recommended dialogue template to update historical dialogue data, and iteratively update and improve the recommended dialogue template database, so that a better intelligent dialogue effect is obtained.

Drawings

The above and other features and advantages of the present application will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

FIG. 1 is a schematic flow chart diagram of a process for generating a recommended speech template database according to one embodiment of the present application.

Fig. 2 is a schematic flow diagram of a method for dialogs based on recommended speech according to an embodiment of the present application.

Fig. 3 is a schematic block diagram of an apparatus for dialogs based on recommended speech according to an embodiment of the present application.

FIG. 4 is a schematic block diagram of an electronic device according to one embodiment of the present application.

Detailed Description

Exemplary embodiments will now be described more fully with reference to the accompanying drawings. The exemplary embodiments, however, may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art. In the drawings, the size of some of the elements may be exaggerated or distorted for clarity. The same reference numerals in the drawings denote the same or similar structures, and thus a detailed description thereof will be omitted.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the application. One skilled in the relevant art will recognize, however, that an embodiment of the application can be practiced without one or more of the specific details, or with other methods, components, etc. In other instances, well-known structures, methods, or operations are not shown or described in detail to avoid obscuring aspects of the application.

The intelligent dialogue scheme based on the dialoging recommendation algorithm can be applied to various intelligent dialogue scenes. For example, in a smart recruitment scenario, a recruiter first sends recruitment information (e.g., a recruitment advertisement, etc.) about a recruitment position to the public or a specific group of candidates, and a human resource specialist (HR) can initially communicate with the candidates using a smart conversation system, such as a conversation robot, before making a phone call or face-to-face conversation or communication with the candidates, to obtain candidate information needed for recruitment and the candidates' needs and opinions of the recruitment position as much as possible. Thus, automated dialog in intelligent recruitment can be used as a means to supplement candidate information and to perform preliminary matching and screening between candidates and the recruitment position. For another example, in a smart customer service scenario such as intelligent handling of repairs and complaints, in order to reduce the working pressure of customer service professionals and improve the accuracy and specificity of the service, an intelligent dialog system may be used to initially communicate with the user and guide the user to accurately describe repair information or complaint content in order to find professionals who can resolve the problem. The work load of customer service personnel can be reduced even by solving the customer service requirement of the user through the intelligent dialogue system. In addition, the intelligent dialogue system can be used for recognizing the emotion of the user, properly meeting the requirements of the user and improving the communication experience of the user.

The intelligent dialog scenarios may include dialog scenarios of chatty type, task type and other dialog types according to their purpose. The conversation scheme based on the recommended speech technology needs to acquire information related to conversation subjects from human conversation participants in a conversation scene as much as possible, and belongs to an intelligent conversation scene of a task type. During a real-time conversation, at least two conversation participants are involved, one being a human conversation participant (e.g., a candidate in a smart recruitment scenario, or a user or customer in a smart customer service scenario), and the other being an intelligent conversation system (e.g., a conversation robot) that generates conversation content based on a recommended conversation template to ask or answer questions of the human conversation participant. The real-time conversation process may also include more participants. In this context, a conversation participant is intended to refer to a human conversation participant in a conversational communication with the intelligent conversation system, i.e., a conversation party of the intelligent conversation system.

The prior information of the intelligent recruitment scenario referred to in the present application refers to information related to the conversation participants and information related to the intelligent conversation scenario. The information about the conversation participants may characterize the attributes of the conversation participants themselves. For example, in the intelligent recruitment scenario, the resume information of the candidate can be the information part related to the conversation participant (candidate) in the prior information, and the recruitment information (such as the recruitment advertisement) is the information part related to the conversation scenario in the prior information. The candidate can have sent the resume to the recruiter prior to the automated conversation process for intelligent recruitment, assuming that the resume data has been sent to the intelligent dialog system. In the intelligent customer service scenario, the user has submitted preliminary information about the fault description and complaint content electronically as a priori information through a questionnaire or template prior to talking with the intelligent dialog system.

At least one of the prior information and historical dialog data of the historical dialog is used to generate a default recommended dialog template database for use by the intelligent dialog system. The a priori information and the historical dialogue data may be of a variety of data types. For example, the prior information such as the candidate resumes can be paper resumes or digitized resume data obtained at a web interface or mobile program (APP) provided by the recruiter. Non-digital information such as paper resume can be converted into digital data through digital processing. It is also possible to convert different types of digitized data into the specific type of digitized data required by the intelligent dialog system. For example, historical conversation data, including conversations between human conversation participants (e.g., telephone or face-to-face interview data between HR and candidates) or conversations between intelligent conversation systems and human conversation participants, is typically recorded voice or video data, and may also be text or image data that processes the voice or video data, and then the graphics, image or voice/video data may be converted to text data for analysis and use by the system.

In the intelligent conversation scene of the task type, a default conversation process can be set according to the scene requirement. In a conversation process, an intelligent conversation system needs to ask questions of human conversation participants to obtain enough information that a conversation task is expected to obtain. Meanwhile, the intelligent conversation system also gives corresponding replies according to the problems proposed by the human conversation participants, so that the peer-to-peer requirement of the human conversation participants for obtaining information is met. Taking the intelligent recruitment scenario as an example, the role of the intelligent dialog system (dialog robot) is equivalent to human HR, and tasks of asking candidates and responding to queries of candidates performed by human HR in the interviewing process need to be completed in the dialog.

During the course of a conversation, the conversation robot may need to converse with a candidate for at least one issue. One or more of the issues in a real-time conversation process is a topic of the conversation in the conversation scene, which may also be referred to as a conversation intent, to indicate the intent or purpose of the conversation. For example, in an intelligent recruitment scenario, a conversation robot is asked for a candidate (conversation participant) for each conversation topic instead of HR or is asked for a candidate (conversation node) for each conversation topic first, the candidate is waited to reply, then the acquisition degree of information desired to be acquired by the conversation topic and the state of the candidate are considered, the question asking for the conversation topic again is continued, or the candidate is waited to ask the conversation robot and give a targeted reply. The concept of a conversation node, which may also be referred to herein as a topic node or an intent node, is a conversation opportunity or link for obtaining information related to a certain conversation topic, which includes one or more questions and corresponding replies from conversation robots and/or candidates corresponding to the conversation topic. Therefore, the number of recommended dialogs templates of question type or answer type usable by the dialog robot in each dialog node is one or more. The plurality of recommended dialogs templates are used for solving the problem that a single question is insufficient to acquire all information related to the conversation subject, or the plurality of recommended dialogs templates are used for solving the problem that when the question or the reply is given to the conversation subject, the conversation content generated by using the recommended dialogs templates is denied by a candidate, and the conversation content needs to be generated again by replacing the recommended dialogs templates.

If all the information expected to be acquired by the current conversation topic is acquired and the candidate does not have a question to ask the conversation robot further, the real-time conversation link for the conversation topic is ended, and the conversation process enters the next conversation topic, that is, the current real-time conversation enters the conversation node of the next conversation topic. Otherwise, the real-time conversation is still in the current conversation node.

As described above, the default conversation process for an intelligent conversation scenario may consist of a conversation node divided into one or more conversation links for one or more conversation topics. The intelligent conversation system (conversation robot) selects a recommended conversation template to generate conversation content to complete the question or the answer to the conversation participants in corresponding conversation nodes according to the real-time conversation data of the current conversation topic and the prior information related to the conversation participants in one or more conversation topics involved in the conversation. And repeatedly executing the conversation link of each conversation topic in each conversation node until all the information expected to be acquired by the conversation topics is obtained or the conversation is terminated by the intelligent conversation system or the conversation participants under the condition that a preset termination condition is met.

According to embodiments of the present application, the dialog process with human dialog participants is typically conducted in a voice or video manner, but techniques for generating dialog content based on conversational skills typically employ algorithms or models based on textual data. Thus, when the conversation robot asks or replies to the participants of the conversation the latter, the generated conversation content (i.e., the historical conversation data or the real-time conversation data from the intelligent conversation system/conversation robot) can be converted into voice data or information. Accordingly, upon obtaining a reply or a question from a human conversation participant (i.e., historical conversation data or real-time conversation data from the conversation participant), the audio/video data may be converted to text data. There is a bi-directional conversion of data format between text data and audio/video data.

The scheme of the application is described as an example of a recruitment conversation process between a phone robot and a human candidate in an intelligent recruitment scene with reference to the attached drawings 1 to 3. Those skilled in the art will appreciate that the present embodiments are illustrative only, and are not limiting upon the scope of the claimed subject matter.

The recommended speech technology-based conversation method firstly combines historical conversation data and/or prior information of historical conversations to generate a default recommended speech technology template database, and then selects a recommended speech technology template based on the real-time conversation data and the prior information in the real-time conversation process to generate conversation contents in real time. Meanwhile, real-time dialogue data of the real-time dialogue process can be returned periodically or when preset conditions are met to perform off-line training and update the default recommended dialect template database, so that closed-loop iterative updating of the recommended dialect is realized.

FIG. 1 illustrates the generation and updating process of a default recommended dialogs template database. Those skilled in the art will appreciate that upon the first use of the conversation robot for a smart recruitment conversation (which may also be referred to as a cold start of the smart conversation system), an initial version of the default recommended speech template database may be generated based on historical conversation data recorded in the past (e.g., between human HR and candidates) that processes audio or video data into a textual form of historical conversation. After intelligent dialogs are conducted by using the recommended dialogs generated by the default recommended dialogs template database, the continuously recorded real-time dialog data of the real-time dialogs can be added into the historical dialog data set as new historical dialog data 101, and the new version of the recommended dialogs template database is obtained through updating.

The process of using the default recommended dialogues template database of the default recommended dialogues generation algorithm mainly includes the steps of extracting the topic of the dialog, the dialog tags and the historical dialog corpus from the historical dialog data S110, determining the slot positions and the corresponding relationships S120, and removing the content items in the historical dialog corpus to generate the default recommended dialogues template database S130.

In step S110, the questions and corresponding responses that the HR and the candidate have proposed in the historical conversations are extracted and summarized using algorithms such as clustering of unsupervised clustering and keyword fishing, so as to obtain the relevant conversation topics in the recruitment conversation scene, the conversation tags related to the conversation topics, and the historical conversation corpora corresponding to the conversation topics.

In a default conversation process, it is common to start with a question from one party of the conversation and then a reply from the other party. Therefore, extracting relevant data and information from (historical or real-time) dialog data requires detecting the type of current dialog content in the historical or real-time dialog data, i.e. whether the dialog content is from a questioning dialog participant (HR or candidate for questioning in the historical dialog, dialog robot or candidate in the real-time dialog of the intelligent dialog process) or from a answering dialog participant (HR or candidate for questioning in the historical dialog, dialog robot or candidate in the real-time dialog of the intelligent dialog process). In this context, the dialog types of a real-time dialog are distinguished by a question type and an answer type. During the dialog, the candidate's question of the HR or of the dialog robot may give a reply including specific information, or may give a positive or negative reply to a question, such as a question of the opinion of the tendency, to indicate whether the candidate agrees with the relevant question from the HR or the dialog robot. Typically, the HR or dialog robot verifies that the information that needs to be confirmed in the positive or non-negative reply received from the candidate indicates that the problem just posed, but that a negative reply indicates that at least one of the information that needs to be confirmed is incorrect or negative by the candidate. Meanwhile, the negative reply may also indicate that the conversation party (candidate) is not interested in or has negative emotion in the conversation topic or the current real-time conversation content of the conversation process. During the human conversation, the HR may also give introductory conversation content such as recruitment information (e.g., work content, salary, work hours, etc.) to the candidate to draw the candidate's attention or reverse their negative emotions for the just negative reply, thereby keeping the conversation to the end of the present conversation. The introductory dialog content typically employs an introductory dialog-type template that remains essentially a revised dialog content for a previous question or reply, and thus the introductory dialog content may be categorized into the same dialog type as a real-time dialog for a previous question or reply, e.g., a dialog content that remains to be considered a question type or an answer type.

When extracting information from historical dialogue data (and real-time dialogue data as described below), it is necessary to separate complete dialogue data given by a questioner or a replying party, which is of a questioning type or a replying type, from recorded continuous voice or video data. The complete dialogue data covered by such complete continuous questions or replies output by the same dialogue participant is called dialogue corpora in the dialogue data as the minimum dialogue unit of the dialogue process. The dialog corpus may include one or more sentences and even one or more speech segments having a plurality of sentences. The dialog corpus can be accurately segmented from the dialog corpus, including continuous speech or video data (and continuous text-type dialog data converted therefrom) by means of associated speech recognition algorithms. For example, a complete corpus of dialogues may be determined by identifying whether the dialog data is output by the same party to the dialog, or identifying the mood words (e.g., query words) and/or punctuation marks (e.g., periods or question marks) in the dialog data. In general, in the interaction between two or more parties to a conversation, a plurality of question sentences continuously output by the same party to a conversation can be divided into the same corpus from the conversation data.

After the segmentation of the dialog corpus is completed, clustering the dialog corpus by using a clustering algorithm such as an unsupervised clustering algorithm, for example, for two different dialog corpus sets of a question type and an answer type, dividing the dialog corpus into classes of different topics, and determining the name of each topic class, i.e., the dialog topic related to the dialog corpus in the class, so as to determine one or more dialog topics related in the historical dialog data and the historical dialog corpus under the dialog topic. This step is used to determine which of the large categories of questions that need to be communicated to the candidates during the conversation and what the topic of each category of questions is. Because two different types of dialogue corpus sets of the question type and the answer type are clustered respectively, the historical dialogue corpora belonging to the same dialogue subject class are divided into the historical dialogue corpus of the question type (question type historical dialogue data) and the historical dialogue corpus of the answer type (answer type historical dialogue data), and the classification mode facilitates the generation and the use of subsequent recommended dialogue templates.

For the intelligent recruitment conversation scene, the clustered conversation topics can comprise work content, salary and welfare, work time, promotion mechanism and the like. Wherein, the salary welfare can be further subdivided into two conversation topics of insurance and salary. In the clustering, the historical dialogue corpora relating to subjects such as four-risk one-fund and five-risk one-fund can be clustered into an insurance dialogue subject class, and the historical dialogue corpora relating to annual salary, monthly salary, hourly salary, bonus and the like can be clustered into a salary dialogue subject class. For example, history dialogue corpora of a question type for asking about work content or asking a candidate for satisfaction with the work content, and history dialogue corpora of an answer type for replying to a candidate's work content question belong to a dialogue subject class of the work content.

For the clustered conversation topics and the historical conversation corpus in the class of each conversation topic, at least one conversation label related to the conversation topic is extracted by using a keyword fishing algorithm. The dialog tags are used to characterize the attributes of the various dimensions of the dialog topic, corresponding to the keywords of the dialog topic or intent. For example, for an insured conversation topic, a five-risk one-fund and a four-risk one-fund can be extracted as corresponding conversation tags. For example, in the conversation topic related to working time, the related conversation label can include a time related keyword and other related high-frequency keywords, etc.

Step S110 converts the clustered and keyword captured historical dialogue data into a historical dialogue corpus composed of historical dialogue corpus sets with corresponding dialogue topics and relevant dialogue labels. The historical dialogue corpus can execute clustering and keyword fishing algorithm again to update dialogue topics, relevant dialogue labels and corresponding historical dialogue corpora after obtaining updated historical dialogue data, so that data support is provided for updating of a default recommended dialogue template database.

Next, in step S120, for the clustering and keyword fishing results, entities related to the dialog tags in the history dialog corpus are extracted and identified using NER (Named Entity Recognition) and RE (relationship extraction) algorithms. The entity indicates the content and location of the slot for removal from the historical dialog corpus for both the question type and the answer type. Further, the slot position can be determined and extracted by the NER, and the corresponding relationship between the slot position and the dialog tag can be determined by the RE algorithm.

The NER algorithm may identify entities in the dialog corpus referred to by dialog tags in the historical dialog corpus (question or answer types). An entity is a content item corresponding to or referred to by a conversation label, which may be text data in the form of a character string. And removing the entities from the historical dialogue corpus to obtain the slot positions corresponding to the dialogue tags in the historical dialogue corpus.

The RE algorithm is used to find the context of the slot in the historical dialog corpus. The context includes a reference relationship and a modifier relationship. For example, for conversation tag a, after the NER algorithm determines the corresponding slots A1 and A2 of conversation tag a in the historical conversation corpus, the RE algorithm finds the context of slots A1 and A2. The reference relationship comprises that A1 refers to A2 or A2 refers to A1, and the modification relationship comprises that A1 modifies A2 or A2 modifies A1. The slot corresponding to the conversation label a can be found through the context determined by the RE algorithm.

Therefore, the slot corresponding to the dialog tag in the dialog corpus can be extracted through the NER, and the correspondence between the slot and the dialog tag is determined through the RE algorithm. According to the embodiment of the application, the slot position is determined and the corresponding relation between the slot position and the conversation label is carried out according to the historical conversation corpora belonging to the same conversation type under the same conversation topic class, so that the recommended conversation template generated according to the slot position has the same conversation topic and the attribute of the conversation type. The screening mechanism only marks the conversation labels in the conversation corpus with the same conversation theme and the same conversation type, and only determines the slot positions corresponding to the conversation labels for the fixed context relationship. For example, for a recommended utterance in a case where a candidate does not accept a work schedule such as shift, the recommended utterance template may be searched only in a historical utterance corpus that does not accept work hours, thereby improving accuracy and efficiency of utterance recommendation.

The determined slot position can be rechecked by a service expert to assist in checking whether the slot position setting in the historical dialogue corpus conforms to the language habit.

In step S130, the corresponding entities (content items) in the historical dialog corpus are removed based on the determined slot positions to form slot positions to be filled, thereby converting the historical dialog corpus into a recommended speech template. In this way, a default recommended dialog template database with slots is generated from the historical dialog corpus. The recommended dialect template database not only comprises recommended dialect templates, but also comprises dialog topics and dialog types corresponding to the recommended dialect templates, dialog tags related to the dialog topics, and corresponding relations between the dialog tags and the slots.

An example of a default recommended dialogs template generated is as follows:

if the candidate gives a negative reply to the question about working hours, i.e., does not accept a working hour regime such as shift, the default recommended dialog template generated may be: "see your work experience associated with \_ _, it is believed that you can also understand that this is almost always the case for the time of the service industry, but we can ensure that the working time of work shift is more flexible, there is just time for some other things to be done. <xnotran> , ____ , ____ . </xnotran> <xnotran> ____ , ____ . </xnotran> But also see that you now also live: \\\ _, our work place is also assigned nearby, reducing commuting time and difficulty, you can consider comprehensively? "where the underlining is the slot corresponding to the conversation tag associated with the conversation topic of the work hour.

If the real-time conversation data from the real-time conversation is to be used as updated data for the historical conversation data 101 in a subsequent process, steps S110 through S130 may be performed again to obtain an updated default recommended conversation template database. According to the embodiment of the application, the updated default recommended speech technology template database not only comprises the updated recommended speech technology template, but also comprises the updated conversation topic after re-clustering, the conversation label related to the updated conversation topic, the updated historical conversation corpus in the conversation topic class, and the updated slot position setting and the corresponding relation between the slot position and the conversation label.

Fig. 2 shows a schematic flow of a method for conducting a dialog based on recommended speech according to an embodiment of the application.

The method flow mainly includes, for a current conversation topic in one or more conversation topics involved in a current real-time conversation, selecting a recommended conversation template based on real-time conversation data and prior information of conversation participants to generate conversation content in real time S210, and acquiring a reply of the conversation participants (e.g., candidates) to the generated conversation content, and updating the real-time conversation data based on the reply to generate next conversation content in real time for an intelligent conversation system (e.g., a conversation robot) S220.

For example, step S210 may ask the candidate at the corresponding dialogue node for dialogue topics such as academic calendar, age, graduation time, etc., respectively, and obtain or confirm information of the candidate academic calendar, age, graduation time, etc. In particular, step S210 first determines a dialog type, such as a question type and an answer type, for the real-time dialog from the current dialog topic in sub-step S211. The process of determining a conversation type from real-time conversation data is similar to the process of determining a conversation type from historical conversation data described above with reference to fig. 1 and will not be described in detail herein.

Next, a dialog tag (also referred to as a real-time dialog tag) related to the real-time dialog of the current dialog topic is extracted from the real-time dialog data and the a priori information in sub-step S212. The conversation tags mainly come from conversation tags (conversation content tags) 201 related to real-time conversation contents and conversation tags (a priori conversation tags) related to a priori information, wherein the conversation tags related to the a priori information further include conversation tags (conversation participant tags) 202 related to conversation participants and conversation tags (conversation scene tags) 203 related to intelligent conversation scenes. These conversation tags combine to form a real-time conversation tag that is associated with the current real-time conversation.

The conversation content tag 201 is a conversation tag extracted from real-time conversation content. The conversation content tags 201 may be conversation tags selected among the conversation tags associated with the current conversation topic, where the topic-related conversation tags have been predetermined by a keyword fishing algorithm in the process of generating the default recommended conversation template database shown in fig. 1. For example, when the dialog topic of the current dialog node is payroll, the dialog tags related to the payroll topic include current payroll, expected payroll, graduation time, and the like. Accordingly, relevant information items such as current salary, expected salary and graduation time can be extracted from the real-time dialog content as the dialog content tag 201.

As described above, in the intelligent recruitment conversation scenario, the conversation participant tags 202 associated with the conversation participants are from the resume information of the candidate, and thus the conversation participant tags 202 may also be referred to as candidate tags. The conversation participant tags 202 can be obtained based on resume information obtained prior to the intelligent recruitment conversation using a resume parsing algorithm, such as clustering. The resume parsing algorithm can extract candidate related information items in the resume information as candidate tags. The resume information may be resume data in a structured form or an unstructured form. Compared with the structured form, the resume parsing algorithm can process the unstructured resume data to generate the structured resume data, and extract information items such as the names, ages, academic calendars, expected work places (e.g. cities and regions), current work places (e.g. cities and regions), expected work categories, current work categories, work experiences and the like of the candidates as candidate tags.

The conversation scenario tag 203 associated with the conversation scenario is then associated with an intelligent conversation scenario, e.g., the conversation scenario tag 203 of the intelligent recruitment conversation is recruitment information. In the intelligent recruitment scenario, the conversation scenario tags 203 related to the conversation scenario are extracted from the previously published recruitment information through a recruitment resolution algorithm such as clustering, and thus can also be referred to as recruitment information tags. If the recruiter has a wide business area and a large distribution of markets (e.g., nationwide businesses), there is a difference in salary and welfare, work content, and work hours for each market region around the country. The recruitment analysis algorithm can extract information most relevant to the conversation topic of the intelligent recruitment conversation from the recruitment information in different areas as a conversation scene label 203. A recruitment information database may be pre-established, with the recruitment information in each recruitment advertisement being from a portion of the recruitment information database, wherein the recruitment information tags may be selected from a set of recruitment information tags in the recruitment information database. The recruitment information tag can include, for example, work content of the post to be recruited, work time, work place, work category, work salary and welfare, and the like.

In practical use, a considerable portion of the recruitment advertisement formats are unified, such as "work content: xxx, on time: xxx ", the recruitment information tag can be obtained by extracting the content item corresponding to the corresponding information item. If the recruitment advertisement is not released in a digital manner, text data about the recruitment information can be acquired by performing text recognition on a unified non-digital recruitment advertisement (such as a recruitment poster in a picture format) through an OCR (optical character recognition) algorithm, and then a related recruitment information tag, namely a conversation scene tag 203, is extracted by using NER and RE algorithms. In fact, when the pre-acquired resume information of the candidate is in a non-digital format, the resume information in the form of text data can be acquired in a similar manner to further extract the candidate tag.

Similar to the conversation content tags 201, the conversation participant tags 202 and the conversation scene tags 203 are generally extracted in the resume information parsing and the recruitment information parsing only for the conversation tags that have been predetermined by the keyword bailing algorithm in the process of generating the default recommended conversation template database shown in fig. 1, with the purpose of trying to acquire information related to a conversation topic focused on by the current conversation node when matching the recommended conversation template using the conversation tags, candidate tags, and recruitment information tags including real-time conversation content.

After the conversation content tag 201, the conversation participant tag 202 and the conversation scene tag 203 are obtained, a data alignment operation may be performed to merge the conversation tags (i.e., the conversation content tag, the conversation participant tag and the conversation scene tag) related to the conversation content, the conversation participants and the conversation scene into a conversation tag (a real-time conversation tag) related to a real-time conversation, so as to obtain the latest and most comprehensive personalized information up to the current real-time conversation. The above-mentioned combination operation does not need to perform clustering because the clustering operation is used to generate a default recommended dialog template database in advance and/or extract candidate tags and recruitment information tags from the resume information and the recruitment information, and only the dialog content tags 201 related to the real-time dialog data of the current dialog in each dialog node are extracted and updated in order to improve the recommendation efficiency and speed during the real-time dialog.

Next, in sub-step S213, a recommended speech template that best matches the conversation requirement or purpose of the current conversation topic is selected from the default recommended speech template database to generate the conversation content to be output by the conversation robot to the human candidate. The current optimal recommended dialogs template may be determined based on how well the real-time conversation tags associated with the real-time conversations match the slots of the plurality of recommended dialogs templates in the default recommended dialogs template database. The matching degree of the correspondence between all the conversation tags in the conversation content tag 201, the conversation participant tag 202 and the conversation scene tag 203 in the personalized information obtained after the data alignment operation and the slots of all the templates in the default recommended conversation template database can be calculated. For example, for all conversation tags in the personalized information of the current conversation, based on the number of slots included in the default recommended conversation template, the ratio of the number of slots having a corresponding relationship with the conversation tags, that is, the number of corresponding slots respectively filled by the conversation tags, to the total number of slots of the recommended conversation template is used as a matching rate or a matching degree to rank all the templates. The recommended dialogs template with the highest matching rate may be selected as the optimal template for the real-time dialog of the current dialog. For example, if the recommended dialoging template 1 may fill 80% of the number of slots and the recommended dialoging template 2 may fill 100% of the number of slots, the recommended dialoging template 2 is selected as the recommended dialoging template for the current real-time conversation, referred to as the first recommended dialoging template or the best recommended dialoging template. In determining the matching rate, the context between the slots may not be used, but may be based only on the correspondence between the conversation tags and the slots.

In sub-step S214, those conversation tags which have a corresponding relationship with the slots in the first recommended dialogs template determined in sub-step S213 are selected from the conversation content tags 201, the conversation participant tags 202 and the conversation scene tags 203 of the personalized information of the candidate and filled in the corresponding slots, and the complete conversation corpus is generated as the conversation content output by the conversation robot to the candidate in the real-time conversation.

In step S220, after generating and outputting the dialog content for the candidate in sub-step S221, the reply to the dialog content by the candidate may be acquired. The reply may be to provide further specific information for the information query in the question, or may be a positive or negative reply or acknowledgement to the information query in the question.

If it is determined in sub-step S222 that the reply of the conversation participant is a negative reply or includes a negative reply (branch "yes"), it is proved that the conversation contents generated in step S210 do not achieve the intended conversational recommendation effect to produce a negative reply of the candidate. The method proceeds to sub-step S224 where another recommended dialog template of the corresponding dialog type is selected from the recommended dialog template database for the current dialog topic in the dialog node based again on the dialog content tag 201, the dialog participant tag 202 and the dialog scene tag 203 to generate the next dialog content.

As described above, a negative reply indicates that at least one of the information to be confirmed is incorrect or that the candidate is not interested in or has a negative emotion with respect to the content of the current conversation process or the real-time conversation content of the current conversation. Such negative replies may be considered real-time conversation data that includes negative or negative information, where negative information is negative or reverse information or data to the information involved in the conversation content just generated. The negative information shows that the dialog content tag 201, which is related to the real-time dialog at least in the personalization information, needs to be updated and changed. In some cases, the information in the question may also relate to a confirmation of the content in the candidate's resume information and/or the recruitment information, and the negative reply may also have an impact on the conversation participant tags 202 and conversation scenario tags 203 of the current real-time conversation.

After the dialog content tag 201, the dialog participant tag 202 and the dialog scene tag 203 in the personalized information are updated, a recommended dialog template that best matches the dialog requirement of the current dialog is selected again from the default recommended dialog template database in sub-step S224 to generate new dialog content to be output to the human candidate by the dialog bot. At this time, the most suitable recommended speech template may be obtained by selecting a recommended speech template having a different matching rate in the previous matching rate ranking as a new recommended speech template, i.e., the second recommended speech template. The second recommended utterance template may have a lower match rate than the first recommended utterance template, e.g., a second highest match rate in the match rate ranking. The new recommended dialogs template may also be selected based on other criteria. According to the embodiment of the application, since the personalized information is updated by the negative reply so as to obtain the new conversation tag set related to the real-time conversation, the matching degree of the corresponding relationship between all conversation tags and the slots of all templates in the default recommended dialogs template database can be re-determined by using the new conversation tag set (for example, the new real-time conversation tag).

After the second recommended dialogs template is determined, new dialog content is generated by populating slots in the second recommended dialogs template with dialog tags in substep S225 and a candidate' S reply to the new dialog content is again obtained in substep S221.

For example, in the dialog theme related to working hours, when the candidate is unsatisfied with the question related to working hours output by the dialog robot and gives a negative response, the dialog content of the newly generated question type may further be:

"do you be dissatisfied with the early shift schedule, the late shift schedule, or the entire shift schedule? "

Candidate replies are retained as tag information associated with the real-time conversation during the data alignment operation. Accordingly, the contents related to several key dialog tags can be analyzed according to the resume information of the candidate, such as: whether the current residential city is consistent with the city to be engaged, whether the expected working city is consistent with the current city to be engaged, whether the catering related working experience exists, and the like. Meanwhile, a specific conversation label related to the information of the on-off time can be found from the recruitment information related to the post delivered by the candidate.

Then, if the city employed by the candidate is the same as the current city of residence, has relevant working experience, and is not satisfied with the early shift, the corresponding conversation content generated based on the second recommended session template may be: "see you have the relevant work experience, believe you can also understand this work time regime. And although our on-duty time ratioEarlier, but also around 6, 7 o 'clock, 3, 4 o' clock, can go back to rest in the morning, and see you live now:shanghai provinceThe method is consistent with the employment of cities, the work places of people are also distributed nearby, the time and the difficulty of the Tong le are reduced, and people can comprehensively consider.

According to the embodiment of the application, a coping strategy for repeatedly receiving negative replies from the candidates in the conversation nodes aiming at the same conversation topic can be set. For example, after receiving negative responses twice consecutively, the robot ends the intelligent dialogue process and then ends the interview.

If the answer of the conversation participant is not judged to be a negative answer in the substep S222 (branch "no"), it is proved that the conversation contents generated in the step S210 have obtained the intended recommendation effect, and the relevant detailed information can be acquired from the answer of the candidate or confirmation of the candidate for the content of the question asked can be obtained. Next, in sub-step S223, the reply of the present conversation content is used as new conversation data to update the real-time conversation data, and the method returns to sub-step S211, and the conversation robot asks the candidate for the current conversation topic or the next conversation topic again.

Therefore, in the real-time conversation in each conversation node, a recommended dialogues determination algorithm is operated aiming at each question link and each reply link, and real-time conversation contents special for candidates are generated.

The method may further include, after each intelligent conversation using the conversation robot, iteratively updating the algorithm model to form a closed-loop update according to positive or negative replies (accepted or not accepted) by the candidate, in sub-step S226, returning a record of real-time conversation data related to the positive replies accepted by the candidate to the generation and update process of the default recommended speech template database shown in fig. 1. The new historical dialogue data 101 formed by the positively replied real-time dialogue data actually corresponds to the verified correct recommended dialogues sample, which can improve the performance of the algorithmic model.

While the intelligent conversation is completed by using the conversation robot, the manual service can be kept. By counting the acceptance degrees of various types of candidates for the dialog contents of the recommended dialogs, for those types of candidates whose acceptance degrees of the dialog contents generated based on the recommended dialogs template do not meet the preset condition, the manual dialog data for the type of candidates can be continuously supplemented, and the default recommended dialogs template database generation algorithm is re-run to update the exclusive template database of the type of candidates. The preset condition may be, for example, an average value in which the degree of acceptance of the conversation contents relating to the respective conversation types, conversation subjects, and the like for a certain type of candidate is lower than the degree of acceptance of the overall recommended speech. Among them, the type of the candidate may include a type of feature division based on whether the current city of residence of the candidate (e.g., a candidate living locally and a candidate living abroad) is due graduate, etc. According to the dialect recommendation results of different candidate types and statistical data of related parameters (such as high-frequency keywords appearing in the dialect recommendation), a specific dialog tag corresponding to the candidate type can be found, and a specific slot in a template corresponding to the specific dialog tag can also be found, so that a dialog corpus database generated based on historical dialog data, the corresponding dialog tag, the slot and the like are updated for the candidate of the type.

The existing dialect recommendation scheme usually uses past dialog data to determine a recommended dialect template even based on a manually constructed dialog template, and the recommended dialect template is not updated after being generated, so that the best matching recommended dialect template cannot be selected in each dialog node according to real-time dialog contents, and a recommended dialect template database cannot be automatically updated based on closed loop.

In contrast, the proposed proposal for carrying out conversation based on recommended dialogs can combine the prior information associated with the conversation participants and the conversation scenes and the real-time conversation data of the current conversation to timely and accurately acquire the conversation requirements and feedback information of the human conversation participants, select a more appropriate recommended dialogs template to ask questions of the conversation participants or reply to questions of the conversation participants, quickly acquire the information required by the conversation tasks from the conversation, and continuously maintain good communication effect with the human conversation participants before the conversation tasks are completed. On the basis of immediately and accurately acquiring the requirements of the interlocutors and responding, the scheme of the application can also return real-time dialogue data periodically according to the dialogue effect of the recommended dialogue template to update historical dialogue data, and iteratively update and improve the recommended dialogue template database, so that a better intelligent dialogue effect is obtained.

Fig. 3 shows an apparatus 300 for dialogs based on recommended speech according to an embodiment of the application. The apparatus 300 may include an interaction unit 310 and a recommendation unit 320.

The interaction unit 310 is used to acquire real-time conversation data of a current conversation between an intelligent conversation system (e.g., a conversation robot) and human conversation participants 301 such as candidates, and output conversation contents generated according to a recommended dialogues algorithm to the conversation participants 301. The interaction unit 310 may also convert dialog content in text data format provided by the dialog bot to the dialog participants 301 into audio/video data audible or visible to the dialog participants 301, and data in audio (e.g. speech) or video (e.g. real-time video chat) format from the dialog participants 301 into text data that can be processed by the recommendation unit 320 of the device 300. The interaction unit 310 may further include a display unit for providing information or prompts related to the real-time conversation to the conversation participant 301, or an input unit for providing other information input means besides voice or video to the conversation participant 301, or the like, which facilitates conversation communication with the human conversation participant 301. The interactive unit 310 may also provide an interactive interface in the form of a web page or App.

The recommending unit 320 is configured to select, for a current conversation topic in the one or more conversation topics included in the current conversation, a best-matching (first) recommended speech template corresponding to the current conversation topic from the default recommended speech template database based on the real-time conversation data and the prior information of the conversation participant 301; generating conversation content based on the selected recommended conversation template; a reply to the generated conversation content is obtained by the conversation participant 301 and the real-time conversation data is updated based on the reply to generate the next conversation content as appropriate. For example, when the reply of the conversation participant 301 is a non-negative reply, it is confirmed that the generated conversation content is accurately suitable. While extracting the information in the real-time conversation data contained in the reply, it is determined whether the information required for the current conversation topic is completely acquired, and it is determined whether to continue asking questions or answering questions asked by the conversation participant 301 or to enter a conversation node corresponding to the next conversation topic in the conversation node of the current conversation topic based on the determination. If a negative reply is received from a conversation participant 301, it may be determined that the conversation content just generated is inaccurate or that the conversation participant 301 is not interested in the real-time conversation content of the current conversation. Accordingly, a (second) recommended conversation template is selected based on the negative information or data contained in the negative reply, generating more accurate and appropriate conversation content. The recommendation unit 320 may also perform the functions of generating a default recommended dialog template database based on historical dialog data as described above in connection with fig. 1, as well as performing other steps and details of generating real-time dialog content as described in connection with fig. 2, and so on.

It should be noted that although in the above detailed description several modules or units of the device and system for dialogs based on recommendation techniques are mentioned, this division is not mandatory. Indeed, the features and functions of two or more modules or units described above may be embodied in one module or unit according to embodiments of the application. Conversely, the features and functions of one module or unit described above may be further divided into embodiments by a plurality of modules or units. The components shown as modules or units may or may not be physical units, i.e. may be located in one place or may also be distributed over a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present application. One of ordinary skill in the art can understand and implement it without inventive effort.

In an exemplary embodiment of the present application, there is further provided a computer-readable storage medium, on which a computer program is stored, the program comprising executable instructions that, when executed by, for example, a processor, may implement the steps of the method for dialogues based on recommended dialogues in any one of the above-mentioned embodiments. In some possible implementations, the various aspects of the present application may also be implemented in the form of a program product comprising program code for causing a terminal device to perform the steps according to various exemplary embodiments of the present application described in the present description of a method for dialogs based on recommended speech, when the program product is run on the terminal device.

A program product for implementing the above method according to an embodiment of the present application may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.

The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The computer readable storage medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable storage medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Program code for carrying out operations of the present application may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., through the internet using an internet service provider).

In an exemplary embodiment of the present application, there is also provided an electronic device that may include a processor, and a memory for storing executable instructions of the processor. Wherein the processor is configured to perform the steps of the method for dialogs based on recommended dialogs in any of the above embodiments via execution of the executable instructions.

As will be appreciated by one skilled in the art, aspects of the present application may be embodied as a system, method or program product. Accordingly, various aspects of the present application may be embodied in the form of: an entirely hardware embodiment, an entirely software embodiment (including firmware, microcode, etc.) or an embodiment combining hardware and software aspects that may all generally be referred to herein as a "circuit," module "or" system.

An electronic device 400 according to this embodiment of the present application is described below with reference to fig. 4. The electronic device 400 shown in fig. 4 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.

As shown in fig. 4, electronic device 400 is in the form of a general purpose computing device. The components of electronic device 400 may include, but are not limited to: at least one processing unit 410, at least one memory unit 420, a bus 430 that connects the various system components (including the memory unit 420 and the processing unit 410), a display unit 440, and the like.

Wherein the storage unit stores program code executable by the processing unit 410 to cause the processing unit 410 to perform the steps according to various exemplary embodiments of the present application described in the method for dialogs based on recommended dialogs of the present specification. For example, the processing unit 410 may perform the steps as shown in fig. 1 and 2.

The memory unit 420 may include readable media in the form of volatile memory units, such as a random access memory unit (RAM) 4201 and/or a cache memory unit 4202, and may further include a read only memory unit (ROM) 4203.

The storage unit 420 may also include a program/utility 4204 having a set (at least one) of program modules 4205, such program modules 4205 including, but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.

Bus 430 may be any bus representing one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

The electronic device 400 may also communicate with one or more external devices 500 (e.g., keyboard, pointing device, bluetooth device, etc.), with one or more devices that enable a user to interact with the electronic device 400, and/or with any devices (e.g., router, modem, etc.) that enable the electronic device 400 to communicate with one or more other computing devices. Such communication may occur through input/output (I/O) interfaces 450. Also, the electronic device 400 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN), and/or a public network, such as the internet) via the network adapter 460. The network adapter 460 may communicate with other modules of the electronic device 400 via the bus 430. It should be understood that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 400, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present application may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which may be a personal computer, a server, or a network device, etc.) to execute the method for conducting a conversation based on the recommended speech technology according to the embodiments of the present application.

Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the application being indicated by the following claims.

Claims

1. A method of conducting a conversation based on recommended speech, comprising:

obtaining a reply from the conversation participant to the generated conversation content, and updating the real-time conversation data based on the reply to generate a next conversation content,

wherein the conversation topic and the at least one recommended dialogs template corresponding to the conversation topic are determined based on historical conversation data for historical conversations.

2. The method of claim 1, wherein selecting a first recommended speech template from at least one recommended speech template corresponding to the current conversation topic based on real-time conversation data and prior information of conversation participants comprises:

determining a dialog type for the real-time dialog based on the real-time dialog data, wherein the dialog type includes a question type and an answer type;

extracting a real-time conversation label related to the real-time conversation from the real-time conversation data and the prior information;

selecting the first recommended dialog template corresponding to the dialog type based on the real-time dialog tag.

3. The method of claim 2, wherein the conversation topic is associated with at least one conversation tag, and wherein extracting the real-time conversation tag associated with the real-time conversation from the real-time conversation data and the prior information comprises:

extracting conversation tags related to conversation content from the real-time conversation data based on the conversation tags related to the current conversation topic to generate conversation content tags;

based on the conversation tags related to the current conversation topic, extracting the conversation tags related to the conversation participants from the prior information to generate conversation participant tags and/or the conversation tags related to the conversation scene to generate conversation scene tags;

combining the conversation content tags, the conversation participant tags, and the conversation scene tags into the real-time conversation tags.

4. The method of claim 3, wherein the recommended dialog template comprises at least one slot to be filled, the slot having a correspondence with the dialog tag, the selecting the first recommended dialog template corresponding to the dialog type based on the real-time dialog tag comprising:

and determining the first recommended speech template based on the matching degree of the real-time conversation tag and the slot position having the corresponding relation with the real-time conversation tag.

5. The method of claim 4, wherein the recommended dialogs template with the highest degree of match is selected as the first recommended dialogs template.

6. The method of claim 4, wherein generating conversation content based on the first recommended conversation template comprises:

populating the slot with the real-time conversation tag corresponding to the slot of the first recommended conversation template to generate the conversation content.

7. The method of claim 1, wherein updating the real-time conversation data based on the reply to generate the next conversation content comprises:

in response to the reply comprising a negative reply, selecting a second recommended speech template corresponding to a conversation type of the real-time conversation from the at least one recommended speech template corresponding to the current conversation topic based on a real-time conversation tag associated with the real-time conversation; and

generating the next dialog content based on the second recommended dialog template,

wherein the second recommended speech template has a lower degree of matching than the first recommended speech template.

8. The method of claim 7, wherein the conversation topic is associated with at least one conversation tag, wherein the real-time conversation tag is extracted from the real-time conversation data and the prior information based on the conversation tag associated with the current conversation topic, wherein the recommended conversation template comprises at least one slot to be filled, wherein the slot has a correspondence with the conversation tag,

selecting a second recommended dialog template corresponding to a dialog type of the real-time dialog comprises:

determining the second recommended dialog template based on a degree of matching of the real-time dialog tag and the slot having the correspondence with the real-time dialog tag, wherein the second recommended dialog template has a lower degree of matching of the slot than the first recommended dialog template; and

generating the next dialog content based on the second recommended dialog template includes:

populating the slot with the real-time conversation tag corresponding to the slot of the second recommended conversation template to generate the next conversation content.

9. The method according to any one of claims 1 to 8, characterized in that the method further comprises: and generating conversation content aiming at the next conversation topic under the condition that the information extracted from the reply meets the requirement of the current conversation topic aiming at the current conversation topic.

10. The method of any one of claims 1 to 8, wherein the historical dialog data comprises question-type historical dialog data and answer-type historical dialog data, and wherein the dialog topic and the at least one recommended dialogs template corresponding to the dialog topic are determined by:

extracting the conversation theme, the conversation label related to the conversation theme and the historical conversation corpus related to the conversation theme from historical conversation data;

determining a content item related to the conversation label in the historical conversation corpus as the slot position, and determining the corresponding relation between the slot position and the conversation label;

removing the content items in the historical dialog corpus to generate a recommended dialog template of question types and answer types, respectively, that are related to the dialog topic.

11. The method according to claim 10, characterized in that the conversation topic and the conversation label are determined by clustering and/or keyword fishing.

12. The method according to any one of claims 1 to 8, wherein the real-time conversation data and/or manual conversation data whose reply is a non-negative reply is used to update the historical conversation data to update the conversation topic and the at least one recommended conversation template corresponding to the conversation topic.

13. The method of claim 1, wherein the conversation scenario comprises a recruitment scenario, the conversation participants comprise candidates, and the prior information comprises resume information and/or recruitment information for the candidates.

14. The method of claim 1, wherein the conversation scenario comprises a customer service scenario, the conversation participants comprise users, and the prior information comprises repair information and/or complaint information submitted by the users.

15. An apparatus for conducting a conversation based on recommended speech, comprising:

the interactive unit is configured to acquire real-time conversation data and output conversation content to a conversation participant;

a recommending unit configured to select, for a current conversation topic in one or more conversation topics involved in a real-time conversation, a first recommended speech template from at least one recommended speech template corresponding to the current conversation topic based on the real-time conversation data and prior information of the conversation participants, wherein the conversation topic is an intention or purpose of the real-time conversation; generating the conversation content based on the first recommended conversation template; and obtaining a reply by the conversation participant to the generated conversation content, and updating the real-time conversation data based on the reply to generate next conversation content,

16. A computer-readable storage medium, on which a computer program is stored, the computer program comprising executable instructions that, when executed by a processor, carry out the method according to any one of claims 1 to 14.

17. An electronic device, comprising:

a processor; and

a memory for storing executable instructions of the processor;

wherein the processor is configured to execute the executable instructions to implement the method of any one of claims 1 to 14.