Automatic interview method and system and computer program product

Abstract

The invention provides an automatic interview method and system and a computer program product, and relates to the technical field of artificial intelligence. The automatic interview method comprises the following steps: acquiring interview task configuration information; wherein the interview task configuration information comprises an interview object circling condition and an interview outline containing one or more interview problems; based on the interview object circling condition, determining a target user to be interviewed and sending an interview request to the target user; in response to the interview request accepted by the target user, performing one or more rounds of conversations with the target user based on the interview outline through a first large language model to obtain a reply statement of the target user; and generating an interview report corresponding to the interview task based on the reply statement of the target user.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an automated interview method, system, and computer program product. Background Art

[0002] User experience research is an important means to help the production team understand product application and assist in product optimization. A product needs to contact a large number of users throughout its life cycle from design and development to operational iteration. Questionnaires and interviews are effective ways to investigate user product experience.

[0003] In order to conduct user experience research efficiently, some embodiments of this specification provide an automated interview method to automatically implement full-process interview management including interview request distribution, intelligent interview dialogue, and interview report generation. Summary of the invention

[0004] One or more embodiments of the present specification provide an automated interview method, comprising: obtaining interview task configuration information; wherein the interview task configuration information includes interviewee selection conditions and an interview outline containing one or more interview questions; based on the interviewee selection conditions, determining a target user to be interviewed and sending an interview request to the target user; in response to the target user accepting the interview request, conducting one or more rounds of conversations with the target user based on the interview outline through a first language model to obtain a reply statement from the target user; and generating an interview report corresponding to the interview task based on the reply statement from the target user.

[0005] According to the method described in some embodiments of the present specification, the target user to be interviewed is determined based on the interviewee selection condition, including: determining a user status query condition and / or a user behavior query condition based on the interviewee selection condition; determining a first user who meets the user status query condition based on a first database; wherein the first database is used to maintain personal basic information and / or historical behavior statistics of one or more first users; determining a first user who meets the user behavior query condition based on a second database; wherein the second database is used to maintain behavior data of one or more first users; determining the target user to be interviewed based on the first user who meets the user status query condition and / or the first user who meets the user behavior query condition.

[0006] According to the method described in some embodiments of this specification, the first user includes a game player; the personal basic information of the first user includes: account identification, game character information, game level; the historical behavior statistics of the first user include one or more of the following information: login time, game time, number of games, consumption status; the behavior data of the first user comes from the behavior log of the first user in the game.

[0007] According to the method described in some embodiments of this specification, the interview task configuration information also includes an interview time condition; the sending of the interview request to the target user also includes: in response to the interview time condition being met at the current moment, sending the interview request to the user terminal corresponding to the target user.

[0008] According to the method described in some embodiments of the present specification, sending the interview request to the target user also includes: obtaining a historical interview identification list of the target user, the historical interview identification list being used to indicate historical interview tasks in which the target user has participated; when it is determined based on the historical interview identification list that the target user has not participated in the current interview task, sending the interview request to the target user; the method also includes: after obtaining a reply statement from the target user, updating the historical interview identification list of the target user based on the current interview task.

[0009] According to the method described in some embodiments of the present specification, the target user to be interviewed is determined based on the interviewee selection conditions and an interview request is sent to the target user, including: determining an interview triggering behavior based on the interviewee selection conditions; when there is behavior data of a first user matching the interview triggering behavior in a third database, determining the first user corresponding to the behavior data as the target user and sending an interview request to the first user; wherein the third database is used to obtain and store in real time the behavior logs of one or more first users in a preset application system.

[0010] According to the method described in some embodiments of the present specification, the interview task configuration information also includes the purpose of the interview, the background knowledge of the interview and the conversation style information, the interview request includes a link to a model conversation interface for conducting a conversation based on a model; in response to the target user accepting the interview request, one or more rounds of conversations are conducted with the target user based on the interview outline through the first large language model to obtain a reply statement from the target user, and further includes: in response to the target user entering the model conversation interface through the link, an interview opening statement is generated through the first large language model based on at least one of the following information: the purpose of the interview, the background knowledge of the interview, and the conversation style information; and the interview opening statement is displayed on the model conversation interface.

[0011] According to the method described in some embodiments of the present specification, the first large language model is used to conduct one or more rounds of dialogue with the target user based on the interview outline, including: calling the first large language model based on the first interview question in the interview outline by the intelligent agent to generate and output an inquiry sentence corresponding to the first interview question; calling the first large language model in response to receiving a first reply sentence of the target user based on the first reply sentence or based on the first reply sentence and the historical sentences of the target user to determine whether the target user has an ending intention; calling the first large language model in response to the judgment result that the target user has an ending intention and outputting the ending sentence; calling the first large language model in response to the judgment result that the target user does not have an ending intention by the intelligent agent to perform follow-up questions or raise the next interview question in the interview outline.

[0012] According to the method described in some embodiments of the present specification, the interview task configuration information also includes customized requirements for follow-up questions; the calling of the first large language model by the intelligent agent in response to the judgment result that the target user has no intention to end the interview to execute follow-up questions or raise the next interview question includes: calling the first large language model by the intelligent agent to generate and output a follow-up question sentence or an inquiry sentence corresponding to the second interview question based on the first interview question, the first reply sentence, the second interview question in the interview outline, and customized requirements for follow-up questions, or based on the first interview question, the first reply sentence, the second interview question in the interview outline, customized requirements for follow-up questions and the historical sentences of the target user; wherein the follow-up question sentence is used to guide the target user to completely reply to the first interview question and / or to guide the target user to further describe the information points contained in the first reply sentence.

[0013] According to the method described in some embodiments of the present specification, the follow-up customized requirement further includes a follow-up information point; the intelligent agent generates a follow-up sentence or an inquiry sentence corresponding to the second interview question and outputs it based on the first interview question, the first reply sentence, the second interview question in the interview outline, and the follow-up customized requirement, or based on the first interview question, the first reply sentence, the second interview question in the interview outline, the follow-up customized requirement, and the historical sentence of the target user, including calling the first large language model to generate a follow-up sentence or an inquiry sentence corresponding to the second interview question and outputting it based on the first interview question, the first reply sentence, the second interview question in the interview outline, and the follow-up customized requirement, or based on the first interview question, the first reply sentence, the second interview question in the interview outline, the follow-up customized requirement, and the historical sentence of the target user. The first language model is called to perform the following processing based on the user's historical sentences: determine whether the first reply sentence contains the follow-up information point; when the first reply sentence contains the follow-up information point, generate and output a first follow-up question sentence; wherein the first follow-up question sentence is used to guide the target user to further describe the follow-up information point contained in the first reply sentence; when the first reply sentence does not contain the follow-up information point, determine whether the first reply sentence completely replies to the first interview question; when the first reply sentence completely replies to the first interview question, generate and output an inquiry sentence corresponding to the second interview question; when the first reply sentence does not completely reply to the first interview question, generate and output a second follow-up question sentence; wherein the second follow-up question sentence is used to guide the target user to completely reply to the first interview question.

[0014] According to the method described in some embodiments of the present specification, in response to the target user accepting the interview request, one or more rounds of conversations are conducted with the target user based on the interview outline through the first large language model, and also includes: calling the first large language model based on the interview outline by the intelligent agent to determine whether the interview questions in the interview outline are all completely answered by the target user, and when the interview questions in the interview outline are all completely answered by the target user, generating and outputting an end statement.

[0015] According to the method described in some embodiments of the present specification, the generating of the interview report corresponding to the interview task based on the reply sentences of the target user includes: processing the reply sentences corresponding to each interview question of the interview task respectively through the second largest language model to obtain one or more user viewpoints; and obtaining the interview report based on one or more user viewpoints corresponding to each interview question.

[0016] According to the method described in some embodiments of the present specification, the reply sentences corresponding to each interview question of the interview task are processed separately by the second largest language model to obtain one or more user viewpoints, including: clustering the reply sentences corresponding to the same interview question by the second largest language model to obtain one or more reply sentence clusters; extracting one or more user viewpoints corresponding to the interview question based on the one or more reply sentence clusters; or, dividing the reply sentences corresponding to the interview question into one or more preset viewpoints corresponding to the same interview question by the second largest language model based on the one or more preset viewpoints; obtaining one or more user viewpoints corresponding to the interview question based on the preset viewpoints corresponding to the reply sentences.

[0017] According to the method described in some embodiments of this specification, the first large language model corresponds to an interview task; wherein the first large language model is fine-tuned based on the interview background knowledge and / or dialogue style information of the corresponding interview task.

[0018] One or more embodiments of the present specification also provide an automated interview system, including: a configuration module, used to obtain interview task configuration information; wherein the interview task configuration information includes interviewee selection conditions and an interview outline containing one or more interview questions; a distribution module, used to determine a target user to be interviewed based on the interviewee selection conditions and send an interview request to the target user; a dialogue module, used to respond to the target user accepting the interview request, conduct one or more rounds of dialogue with the target user based on the interview outline through a first language model to obtain a reply statement from the target user; and a report generation module, used to generate an interview report corresponding to the interview task based on the reply statement from the target user.

[0019] One or more embodiments of the present specification further provide a computer program product, including computer codes, and when at least part of the computer codes is executed by a processor, any of the aforementioned methods can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] This specification will be further explained by way of exemplary embodiments, which will be described in detail by way of the accompanying drawings. The same numbers in the drawings represent the same structures or steps.

[0021] Figure 1 It is a schematic diagram of an automated interview application scenario according to some embodiments of this specification.

[0022] Figure 2 is an exemplary flow chart of an automated interview method according to some embodiments of the present specification.

[0023] Figure 3is an exemplary flow chart of an interview dialogue according to some embodiments of the present specification.

[0024] Figure 4 is an exemplary flow chart of step 340 according to some embodiments of the present specification.

[0025] Figure 5 It is an exemplary frame of the automated interview system according to some embodiments of the present specification. DETAILED DESCRIPTION

[0026] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the embodiments will be described in detail below with reference to the accompanying drawings. Obviously, the contents described below are some examples or embodiments of this specification. For ordinary technicians in this field, without paying creative work, the technical solutions or means disclosed in this specification can also be applied to other scenarios based on these technical contents.

[0027] It should be understood that the "system", "device", "unit" and / or "module" used in this specification is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

[0028] Unless otherwise specified, technical terms used in this specification to describe components, elements, etc. do not refer to the singular, but may also include the plural. Generally speaking, terms such as "include", "comprise", etc. only indicate that the steps, elements, or components that have been clearly identified are included, and these steps, elements, and components do not constitute an exclusive list, such as the method or device described may also include other steps or components.

[0029] Flowcharts are used in this specification to illustrate the operation steps performed by the device or system of the relevant embodiments, but unless otherwise specified, the order used to describe these steps should not be understood as a limitation on the order in which the steps are performed. A person of ordinary skill in the art can adjust the order in which these steps are performed based on the knowledge and information conveyed by the embodiments of this specification, and the adjustment includes but is not limited to swapping the order, merging multiple steps, and splitting a certain step.

[0030] User experience research is an important means to help the production team understand product application and assist in product tuning. A product needs to contact a large number of users throughout its life cycle from design and development to operational iteration. Questionnaires and interviews are effective ways to investigate user product experience. Among them, the questionnaire includes several preset survey questions, and the user's experience of using the product is collected through the user's one-way response to these survey questions. Interviews can understand the user's experience of using the product by talking or talking with the user, such as staff interviewing and talking with users through instant messaging tools, telephone, face-to-face, etc. Interviews are more flexible than questionnaires and can provide a deeper understanding of user experience.

[0031] However, current interviews rely on manual labor, so the interview efficiency is low. For example, it is difficult for staff to conduct interviews with multiple users at the same time. For another example, some user experience surveys have high effectiveness. If the staff cannot complete interviews with as many users as possible after the user experiences the product due to the large number of work tasks, it will be difficult to obtain more accurate user experience feedback. For another example, after completing the research interview, the staff also needs to spend a lot of time and energy to organize the conversation records and form a research report, which may lead to the generation of the research report not being timely, thus affecting the product optimization process.

[0032] To this end, some embodiments of this specification propose an automated interview method to automatically implement full-process interview management including interview request distribution, intelligent interview dialogue, and interview report generation.

[0033] Figure 1 Schematic diagram of an automated interview application scenario according to some embodiments of this specification. Figure 1 As shown, the application scenario 100 may include a processing device 110 , a network 120 , a storage medium 130 , a first user terminal 140 , and a second user terminal 150 .

[0034] In some embodiments, the processing device 110 may include one or more processors or processing engines (e.g., a single-chip processing engine or a multi-chip processing engine). The processor or processing engine may execute computer instructions to implement the automated interview method shown in one or more embodiments of this specification. By way of example only, the processor or processing engine may include a central processing unit (CPU), an application specific integrated circuit (ASIC), an application specific instruction processor (ASIP), a graphics processing unit (GPU), a physical processing unit (PPU), a digital signal processor (DSP), a field programmable gate array (FPGA), a programmable logic circuit (PLD), a controller, a microcontroller unit, a reduced instruction set computer (RISC), a microprocessor, etc., or any combination thereof.

[0035] The network 120 may include any suitable network that can facilitate information and / or data exchange between components in the application scenario 100. In some embodiments, one or more components in the scenario 100 (e.g., the processing device 110, the storage medium 130, etc.) may exchange information and / or data through the network 120. For example, the processing device 110 obtains data in the storage medium 130 through the network 120. For another example, the processing device 110 may distribute the interview request to the target user (e.g., the first user to be interviewed) through the network 120. For another example, the processing device 110 may interact with the first user terminal 140 through the network 120 to complete the interview dialogue with the target user and obtain the target user's reply statement. For another example, the processing device 110 may push relevant information or an interview report of the interview dialogue to the second user terminal 150 through the network 120. In some embodiments, the network 120 may be any one or more of a wired network or a wireless network. For example, the network 120 may include a cable network, an optical fiber network, a telecommunications network, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network (ZigBee), a near field communication (NFC), an in-device bus, an in-device line, a cable connection, or any combination thereof. The network connection between the various parts may be in one of the above-mentioned ways, or may be in multiple ways. In some embodiments, the network may be a point-to-point, shared, centralized, or other topological structure or a combination of multiple topological structures.

[0036] The storage medium 130 can be used to store data and / or instructions related to the application scenario 100. In some embodiments, the storage medium 130 can store data and / or information obtained from the first user terminal 140, etc., such as the first user's reply statement to the interview question. In some embodiments, the storage medium 130 may include one or more storage components, each of which may be an independent device or a part of other devices. In some embodiments, the storage medium 130 may be set in the processing device 110. In some embodiments, the storage medium 130 may include a random access memory (RAM), a read-only memory (ROM), a large capacity memory, a removable memory, a volatile read-write memory, etc. or any combination thereof. Exemplarily, the large capacity storage may include a magnetic disk, an optical disk, a solid-state disk, etc. In some embodiments, the storage medium 130 may be implemented on a cloud platform.

[0037] The first user terminal 140 may be a terminal (or user terminal) used or owned by the first user. In some embodiments, the first user may be understood as a consumer or user of the product, or may be referred to as a C-end user. Among them, the product may be a physical commodity that can be sold and circulated in the market, such as clothes, food, computers, daily necessities, etc.; the product may also be a service product such as entertainment services, medical services, financial services, etc. The terminal may be a terminal device such as a smart phone, a tablet computer, a laptop computer, a touch screen, a personal computer (PC, Personal Computer), a personal digital assistant (Personal Digital Assistant, PDA), etc. In some embodiments, the terminal may also include a client (or a front-end application), which may be a service platform client, a web browser client, an instant messaging client, etc., wherein the aforementioned service platform (or application system) may be a game service platform, an e-commerce service platform, etc. In some embodiments, the terminal may also include a first client of the interview management service platform provided in some embodiments of this specification. The first user may use the first user terminal 140 to purchase goods or services through the network 120, may also comment on the purchased goods or services through the first user terminal 140, and may also complete the interview dialogue through the first user terminal 140. In some embodiments, the first user terminal 140 may also transmit relevant information and data through the network 120 and other components in the application scenario 100 (eg, the storage medium 130 ).

[0038] The second user terminal 150 may be a terminal used or owned by another user group. The user may be a user who uses the interview management service provided in some embodiments of this specification, such as a product production team or a staff member thereof who automatically completes the distribution of interview requests, intelligent interviews, or obtains interview reports through the interview management service provided in this specification. In order to distinguish, this specification refers to such users as second users, or B-end users. Taking game products as an example, the aforementioned first user may be a game player, and the second user may be a game merchant or its staff, such as a game operator, a game planner, an engineer, etc. Taking physical goods as an example, the first user may be a consumer of physical goods, and the second user may be a merchant or its staff who produces physical goods, such as a product manager, a designer, etc. Similar to the first user terminal 140, the terminal used or owned by the second user may be a terminal device such as a smart phone, a tablet computer, a laptop, a touch screen, a personal computer (PC, Personal Computer), a personal digital assistant (Personal Digital Assistant, PDA), etc. In some embodiments, the terminal of the second user may also include a client, which may be a service platform client, a web browser client, an instant messaging client, etc. In some embodiments, the terminal may also include a second client of the interview management service platform provided in some embodiments of this specification. The second user can use the second user terminal 150 to provide goods or services through the network 120, and can also input interview task configuration information, obtain the first user's interview dialogue or check the interview report through the second user terminal 150. In some embodiments, the second user terminal 150 can also transmit relevant information and data through the network 120 and other components in the application scenario 100 (for example, the storage medium 130).

[0039] For the purpose of intuitive understanding, some embodiments of this specification mainly use the automated interview process of game products as an example to illustrate, but this should not be understood as limiting the scope of this specification. The automated interview method disclosed in the embodiments of this specification can be applied to the user experience research process of any product, and can even be applied to various types of interviews such as academic and current affairs.

[0040] Figure 2 is an exemplary flow chart of an automated interview method according to some embodiments of this specification. In some embodiments, Figure 2 The process 200 shown may be performed by a processing device (such as Figure 1 The process 200 may be performed by the processing device 110 shown in the figure. As an example, the processing device may be a server of the interview management service platform shown in some embodiments of this specification. In some embodiments, the process 200 may be implemented by an automated interview system 500 deployed on the processing device. Figure 2As shown, process 200 may include the following steps.

[0041] Step 210 , obtaining interview task configuration information. In some embodiments, step 210 may be implemented by configuration module 510 .

[0042] In some embodiments, the second user may input interview task configuration information to the processing device through the second user terminal to describe or specify the interview task. The interview task configuration information may describe the interview task from the dimensions of one or more information items. In some embodiments, the interview task configuration information may include interviewee selection criteria and an interview outline including one or more interview questions.

[0043] The interviewee selection condition is used to describe which interviewees need to be interviewed for the interview task. In some embodiments, the interviewee selection condition can be determined from aspects such as user status and user behavior.

[0044] The user status may include relatively static attributes or features such as the user's gender, occupation, age, habitual preferences, etc. Taking a game application as an example, the user status may include the first user's basic personal information, such as gender, age, account identification, game character information, game level, etc. As an example, the account identification may be the first user's game ID, character ID, game account, etc., and the game character information may be the relevant features of the virtual character controlled by the first user in the game, for example, the game character information may be the character type, etc. The user status may also include the first user's historical statistical information, such as the first user's login time, game time, number of games, consumption status, etc. Among them, the login time may be the statistical value of the time the first user logs in to the game platform through his game account, the game time may be the statistical value of the time the first user logs in to the game platform and plays the game, the number of games may be the number of times the first user participates in the game, and the consumption status may be the first user's consumption (such as purchasing equipment, skins, skills, etc.) level or consumption amount interval value in the game platform. It can be understood that historical statistical information may be the statistical result of the user's historical behavior to a certain extent. Although it involves user behavior, historical statistical information focuses on the description of the user's habitual preferences.

[0045] In some embodiments, the interviewee selection criteria may be determined based on the user status. As an example, it may be the first user whose level is greater than 10, or the first user whose game time is greater than 500 hours.

[0046] User behavior can be understood as a relatively dynamic feature, which can be specifically the behavior of the first user in a service platform. Continuing with the example of a game application, user behavior can be one or more specific behaviors of the first user in the game platform, for example, the user behavior can be the purchase of a certain type of equipment, the completion of a certain game task, the initiation of a battle, the login to the game, the addition of friends, etc.

[0047] In some embodiments, the interviewee selection criteria may be determined from user behavior. For example, it may be the first user who has completed a certain game task.

[0048] The interview outline is used to record the interview questions that need to be communicated or asked to the first user during the interview, which can reflect the main content that needs to be investigated or understood from the first user in the interview task. As an example, the interview question may be "What is your overall feeling about playing 'xx game' today?", and for another example, the interview question may be "Which aspect of the game design do you like most?". In some optional embodiments, the interview outline may also specify the opening and / or closing remarks during the interview dialogue. Exemplarily, the opening remarks may be "Hello, thank you very much for participating in our interview.", and the closing remarks may be "Thank you for your patience in answering, and I wish you a happy life!".

[0049] In some embodiments, the interview task configuration information may also include one or more of the following information: interview time conditions, interview purpose, interview background knowledge, dialogue style information, follow-up customization requirements, etc. The interview time conditions may specify the initiation time of the interview request related to the interview task, and the interview purpose may be a description of the purpose that the interview task hopes to achieve, such as the interview purpose may be to understand the first user's evaluation of the xx game. The interview background knowledge may be basic knowledge related to the interview task, such as the game story background, the game gameplay, the professional terms involved in the game, the details of the recent activities of the game, etc. The dialogue style information may be the communication method or characteristics used in the interview dialogue process, such as the business meeting style, the academic discussion style, the humorous and relaxed style, the teaching and puzzle-solving style, the emotional care style, etc. The follow-up customization requirements may be the requirements for the large language model to respond to the target user during the interview dialogue process. For example, the follow-up customization requirements may include completing the interview purpose, politely refusing to accept the reply sentence that deviates from the interview purpose or the interview outline too far, and if the reply sentence contains xx information points, giving priority to these information points.

[0050] In some embodiments, the interview task can correspond to the interview task configuration information. The second user can create an interview task by inputting the interview task configuration information into the processing device or the interview management service platform. After the interview task is created, the interview management service platform can execute the subsequent steps by itself and automatically complete the corresponding interview process.

[0051] Step 220 , based on the interviewee selection criteria, determine the target user to be interviewed and send an interview request to the target user. In some embodiments, step 220 may be implemented by the distribution module 520 .

[0052] In some embodiments, the processing device or distribution module 520 can obtain one or more query conditions based on the interviewee selection conditions, and perform a query based on the query conditions to determine one or more first users who meet the interviewee selection conditions and use them as target users to be interviewed.

[0053] In some embodiments, the distribution module 520 can obtain the user status query condition and / or the user behavior query condition based on the interviewee selection condition. As an example, the user status query condition may include the first user whose level is greater than 10, and the user behavior query condition may include the first user who has completed the xx game task.

[0054] In some embodiments, a query can be performed in a database based on a query condition to determine the target user. The database can be a data system that stores or maintains data related to the first user. In some embodiments, the account information and / or historical statistical information of the first user can be maintained through the first database, and the first user who meets the user status query condition can be determined based on the first database. The relevant description of the account information or historical statistical information can be found in the previous text and will not be repeated here. The first database can be a database with relatively low requirements for data update or processing timeliness, which can be used to store the basic information, account information, historical data of the first user or historical statistical information obtained by statistics of historical data. In some embodiments, the data in the first database can be manually imported or automatically pushed from a service platform. As an example only, the first database can establish a communication connection with a service platform and regularly obtain and store the relevant data of the first user from the service platform. In some embodiments, the first database can be implemented based on hive technology. Hive is a data warehouse tool based on Hadoop, which is used for data extraction, conversion, and loading. It is a mechanism that can store, query and analyze large-scale data stored in Hadoop. The hive data warehouse tool can map structured data files to a database table, and provide SQL query functions, which can convert SQL statements into MapReduce tasks for execution. In some embodiments, an SQL statement for user status query can be obtained based on the interviewee selection conditions, and a first user who meets the conditions can be determined from the first database based on the SQL statement.

[0055] In some embodiments, the database may also be a second database that stores or maintains the behavior data of one or more first users, and determines the first user who meets the user behavior query condition based on the second database. The relevant description of the behavior data can be found in the previous text and will not be repeated here. The second database may be a database with high data update or processing timeliness, which can receive and store the behavior data of the first user on a service platform (such as a game platform) in real time. As an example, the second database can establish a communication connection with a service platform, and the service platform can generate a behavior log in real time according to the behavior of the first user, and synchronize the data in the behavior log to the second database in a timely manner. It is understandable that the behavior data of the first user in the second database can come from the behavior log of the user. In some embodiments, real-time or timely can be understood as data, information or operation can be processed, transmitted or responded immediately after it occurs, but in actual applications, according to the specific business scenario, real-time or timely may have different degrees of short-term delays. As an example, the delay can be hour-level, graded, second-level, millisecond-level, etc., but generally the delay will not exceed one day. In some embodiments, the second database can be implemented based on the kafka message system. Kafka is a message middleware that can be used to store messages. It can be understood as an open source stream processing platform that supports high throughput and can efficiently process the action stream data of the first user in the service platform based on the flink computing framework.

[0056] In some embodiments, the target user to be interviewed can be determined based on the first user who meets the user status query condition and / or the first user who meets the user behavior query condition. As an example, the first user who meets the user status query condition can be used as the target user, or the first user who meets the user behavior query condition can be used as the target user, or the two types of first users can be combined as the target user. Some embodiments of this specification provide more adaptive management or maintenance of user status data and user behavior data based on database technology, and determine the target user based on the interviewee selection condition, so as to efficiently realize the distribution of interviews with precise groups of people.

[0057] After determining the target user, the distribution module 520 can send an interview request to the user terminal (such as the first user terminal) of the target user. In some embodiments, the interview request may include a link to the model dialogue interface. By clicking the link, the model dialogue interface can be presented on the terminal. The model dialogue interface can refer to an interface for conducting a dialogue based on a model, and the target user can conduct an interview dialogue with the large language model through the interface. In some embodiments, the interview request can also include an interview task identifier, which can specifically be a number consisting of letters or numbers, or the name of the interview task. In some embodiments, the interview request can be sent to the target user via SMS, email, or an instant dialogue interface within the service platform.

[0058] In some embodiments, an interview request can be sent based on the interview time condition in the interview task configuration information. Specifically, the processing device or distribution module 520 can obtain the current moment and determine whether the current moment meets the interview time condition. When it meets the condition, an interview request is sent to the user terminal corresponding to the target user. In some embodiments, the interview time condition can be the time point at which the interview request is started to be sent, or it can be the time period for sending the interview request. The interview time condition can be set based on business needs, for example, it can be set based on the launch time of a certain product, and for another example, it can be set based on the time of holding a certain event. As an example, the interview time condition can be 00:00 on August 30, 2024. When the current moment is greater than this moment, an interview request is sent to the target user. For another example, the interview time condition can be from August 25, 2024 to August 27, 2024. When the current moment is within this period, interview requests are sent to the target user one after another. By setting the interview time condition, the interview timing can be accurately controlled as needed, making product research and other work more timely.

[0059] In some embodiments, the interview triggering behavior can also be determined based on the interviewee selection conditions. Specifically, the interview triggering behavior can be determined based on the description of the user behavior conditions in the interviewee selection conditions. As an example, when the interviewee selection conditions include completing the xx game task, the completion of the xx game task can be determined as the interview triggering behavior. Further, when there is behavior data of the first user matching the interview triggering behavior in the third database, the first user can be determined as the target user and an interview request can be sent to the first user. The third database can be a database that acquires and stores the behavior logs of one or more first users in a preset service platform in real time. In some embodiments, the third database can be built in a similar way to the second database, or the third database can be the same database as the aforementioned second database. As an example, when the behavior data of a first user completing the xx game task is received in the third database, the relevant information of the first user (such as user identification or contact information) can be pushed to the processing device or distribution module 520 in a timely manner, and the processing device or distribution module 520 can send an interview request to the user terminal of the first user in a timely manner. This embodiment can determine the target user in real time according to the interviewee selection conditions and send the interview, further improving the timeliness of the survey and ensuring that the interview can be completed in time at the best time.

[0060] In order to prevent the interview request of the same interview task from being sent to the same target user multiple times, thereby bringing a poor user experience. For example, when the triggering interview behavior is to complete the xx game task, when the first user A completes the task multiple times, the distribution of the interview request will be triggered, which will repeatedly disturb the first user A and bring a poor user experience. In some embodiments, when sending an interview request to the target user, a historical interview identification list of the target user can be obtained, and the historical interview identification list is used to indicate the historical interview tasks in which the target user has participated. Based on the historical interview identification list, it can be determined whether the target user has participated in the current interview task. When the target user has not participated, the distribution module 520 sends the interview request to the target user. In some embodiments, the historical interview identification can be an identification of the interview task in which the corresponding target user has participated in history, which is maintained in the form of a list (or array), which can be stored in the first database together with the user status data corresponding to the first user, and when the first user is determined as the target user, it is sent to the distribution module 520 together with the identification or contact information of the user. Alternatively, the historical interview identifiers of the first user may be stored in a storage system with a faster reading speed (such as Redis). When sending an interview request to the target user, the distribution module 520 first reads the historical interview identifier of the user from Redis, and sends the interview request to the target user when it is determined that the user has not participated in the current interview task. In some embodiments, the historical interview identifier list of the target user may be updated after the target user completes an interview task, such as by adding the completed interview task identifier as a list element to the historical interview identifier list. More information about updating the historical interview identifier list can be found in the following text and will not be repeated here.

[0061] Step 230 , in response to the target user accepting the interview request, one or more rounds of dialogue are conducted with the target user based on the interview outline through the first language model to obtain a reply statement from the target user. In some embodiments, step 230 may be implemented by the dialogue module 530 .

[0062] In some embodiments, when the target user clicks on a link in the interview request, it can be determined that the target user has accepted the interview request. In response to the target user clicking on the link, a model dialogue interface can be displayed on the user terminal of the target user (such as the first user terminal 140). In some embodiments, the model dialogue interface can include a display window and an input control, the display window is used to display the dialogue content of the dialogue participants (such as the large language model and the target user), and the input control is used to enter the dialogue content of the target user under the triggering of the target user. As an example, the input control can be a text input box, in which the target user can enter the dialogue text, or the input control can also include a voice entry control, and the target user can enter the dialogue content in voice form by triggering the voice entry control. In some embodiments, the input control can also support the input of content such as pictures and videos.

[0063] In some embodiments, in response to the target user accepting the interview request (such as the target user accessing the model dialogue interface by clicking a link in the interview request), the first large language model generates an interview opening statement, and the statement is displayed in the model dialogue interface. A large language model (LLM) refers to a machine learning model that can perform complex natural language processing tasks through large-scale text data training, and its parameter quantity is generally in the order of millions to billions. As an example, the large language model can specifically be GPT, Megatron-LM, LaMDA, LlaMa, vicuna, Baichuan, Qianwen, etc. It should be understood that "first", "second", etc. in some embodiments of the specification are mainly used for distinction, and do not have a substantial limitation on their rhetorical terms. For example, the first large language model is essentially a large language model, and the "first" therein is mainly used to distinguish from another large language model.

[0064] In some embodiments, the first language model can directly output the opening words in the interview outline as the interview opening sentence. In other embodiments, the first language model can generate an interview opening sentence based on one or more of the interview purpose, interview background knowledge, and conversation style information in the interview task configuration information. More information about the interview purpose, interview background knowledge, and conversation style information can be found in the previous text and will not be repeated here. As an example, a prompt word can be generated based on the interview purpose, interview background knowledge, and conversation style information, and the prompt word is input into the first language model to obtain the interview opening sentence generated by the first language model. By generating the interview opening sentence through the interview purpose, interview background knowledge, and conversation style information, the conversation opening can be made more natural and fit the interview task. In some embodiments, the interview task has a corresponding relationship with the first language model. For example, a first language model can be configured for each interview task, and the first language model can be fine-tuned based on the interview background knowledge and one or more of the conversation style of the interview task. After fine-tuning, it is only necessary to indicate in the prompt word that the opening words are generated based on the interview purpose, and the first language model can generate an appropriate interview opening sentence. As an example, the interview background knowledge and sentences with corresponding conversational styles can be used as training corpus, and the first language model can be fine-tuned based on the Fine-tuning interface provided by LoRa, P-Tunning, Open AI, etc.

[0065] In some embodiments, the interview dialogue may be formally entered after the first language model outputs the interview opening statement. Figure 3 is an exemplary flow chart of an interview dialogue according to some embodiments of this specification. In some embodiments, the jump rules can be pre-configured to achieve Figure 3 In other embodiments, the interview dialogue can be managed by an intelligent agent to achieve Figure 3 The process shown. An agent may refer to an agent that can perceive the environment and take actions to achieve specific goals. In some embodiments of the present specification, the aforementioned environment may be the context of an interview dialogue, the action may refer to controls such as the jump of the interview dialogue process, and the goal may refer to obtaining the first user response statement corresponding to the interview question recorded in the interview outline or other subtasks in the interview task. An agent may be a system of software, hardware, or a combination of both, with autonomy, adaptability, and interaction capabilities. The agent perceives changes in the environment, makes judgments and decisions based on the knowledge and algorithms it has learned, and then performs actions to affect the environment or achieve predetermined goals. In some embodiments, the agent can be implemented based on a machine learning model. Figure 3 As shown, the interview dialogue can mainly include the following steps.

[0066] Step 310: Call the first language model to generate and output a query sentence corresponding to the first interview question.

[0067] In some embodiments, the first large language model can be called by the agent based on the first interview question in the interview outline to generate and output an inquiry sentence corresponding to the first interview question. The first interview question can be any interview question in the interview outline. Exemplarily, the first interview question can be the first question in the interview outline. Specifically, one of the interview questions can be obtained from the interview outline by the agent as the first interview question, and the prompt word can be constructed by the agent based on the first interview question to instruct the first large language model to generate a corresponding inquiry sentence. In some embodiments, the first large language model can directly output the first interview question as an inquiry sentence, or the first large language model can improve or adjust the first interview question based on the interview background knowledge, dialogue style information, etc. to generate an inquiry sentence. In some embodiments, if the first large language model has not been fine-tuned, the agent can generate prompt words based on the first interview question and one or more of the interview purpose, interview background knowledge, and dialogue style information, and input them into the first large language model to obtain a corresponding inquiry sentence. If the first large language model has been fine-tuned, the agent can generate prompt words based only on the first interview question, and input them into the first large language model to obtain a corresponding inquiry sentence.

[0068] In some embodiments, the inquiry statement corresponding to the first interview question generated by the first large language model can be displayed in the model dialogue interface for the target user to view.

[0069] Step 320, in response to receiving the first reply statement of the target user, calling the first large language model to determine whether the target user has an ending intention based on the first reply statement or based on the first reply statement and the target user's historical statements.

[0070] In some embodiments, the target user may input a first reply statement for the aforementioned inquiry statement. The first reply statement input by the target user may be a reply to the inquiry statement, or may be a statement of other semantics that is not highly relevant to the first interview question. As an example, the first reply statement may be a statement related to other topics, or may be a statement in which the target user expresses that he does not want to continue the interview. In some embodiments, the first reply statement may be stored (such as recorded in the storage medium 130), and the first large language model may be called by the agent to determine whether the user has an obvious intention to end the interview based on the reply statement of the target user. Specifically, the agent may generate a prompt word based on the first reply statement or based on the first reply statement and the historical statements of the target user before it, and output it to the first large language model, indicating that the first large language model determines whether the target user has an intention to end the interview. In other embodiments, the prompt word constructed by the agent may also include the first interview question, so that the first large language model may also determine whether the target user has an intention to end the interview based on the degree of relevance between the reply statement of the target user and the first interview question.

[0071] In some embodiments, step 330 in process 300 may be executed when it is determined that the target user has an intention to end, and step 340 in process 300 may be executed when it is determined that the target user does not have an intention to end.

[0072] Step 330: Call the first language model to generate and output a closing sentence.

[0073] In some embodiments, the agent can respond to the judgment result that the target user has the intention to end the interview by calling the first language model to generate and output an ending statement. Specifically, the agent can generate a prompt word to instruct the first language model to generate an ending statement based on the ending words in the interview outline. In some embodiments, the agent can also add dialogue style information to the prompt word to indicate that the first language model can generate an ending statement with a specified dialogue style. In some embodiments, the first language model can be fine-tuned to directly improve or adjust the ending words in the interview outline to generate an ending statement that conforms to the current interview dialogue style. Further, the ending statement can be displayed through the model dialogue interface.

[0074] Step 340: Call the first language model to perform follow-up questions or raise the next interview question in the interview outline.

[0075] In some embodiments, the agent can call the first language model to perform follow-up questions or raise the next interview question in the interview outline in response to the judgment result that the target user does not have the intention to end. In some embodiments, when the target user does not have the intention to end, the agent can call the first language model to generate a follow-up question or generate an inquiry statement corresponding to the next interview question (such as the second interview question) in the interview outline based on the judgment result of whether the first reply statement completely replies to the first interview question. Among them, the second interview question can be other interview questions in the interview outline except the first interview question. Specifically, the second interview question can be a question that is later than the first interview question in the interview outline in terms of time sequence or logic. In some embodiments, the agent can also call the first language model to ask questions about the content in the first reply statement. In this way, the agent can flexibly transition to the next interview question naturally based on the reply statement of the target user, or ask questions to the target user to obtain a complete reply to the current interview and conduct in-depth communication on the reply content of the target user. In some embodiments, the agent can call the first language model to perform follow-up questions based on the customized requirements for follow-up questions, so that the follow-up questions can be carried out in an orderly manner around the purpose of the interview. More information about the customized follow-up requirements can be found in the previous article and will not be repeated here. Furthermore, the customized follow-up requirements may include preset follow-up information points. Exemplarily, the follow-up information points may include keywords such as error, bug, and freeze. When the target user's reply statement contains a follow-up information point or content with similar semantics, follow-up questions can be asked about these contents.

[0076] In some embodiments, the first large language model can be called by the agent based on the first interview question, the first reply statement, the second interview question in the interview outline, and the follow-up customized requirements to generate a follow-up sentence or an inquiry sentence corresponding to the second interview question and output it. Among them, the follow-up sentence can be specifically used to guide the target user to fully reply to the first interview question and / or to guide the target user to further describe the information points contained in the first reply sentence. Specifically, the agent can generate a prompt word based on the first interview question, the first reply statement, the second interview question, and the follow-up customized requirements, and input the prompt word into the first large language model to generate a follow-up sentence or an inquiry sentence. In some embodiments, the prompt word can also include the historical sentence of the target user, so that the first large language model can further combine the context of the interview dialogue to conduct follow-up questions or inquiries on the next interview question. In other embodiments, if the first large language model has not been fine-tuned, the agent can also add interview background knowledge, dialogue style information, etc. to the aforementioned prompt word, so that the follow-up sentence or the inquiry sentence is more complete, specific, or more in line with the product positioning.

[0077] In some embodiments, the follow-up question generated by the first large language model or the interrogation sentence corresponding to the second interview question may be displayed in the model dialogue interface so as to be presented to the target user.

[0078] Figure 4 is an exemplary flow chart of step 340 according to some embodiments of this specification. In some embodiments, the agent can construct prompt words based on the above information so that the first language model is as follows: Figure 4 The process shown in the figure performs questioning or inquiry. Figure 4 As shown, step 340 may further include the following steps.

[0079] Step 341, determining whether the first reply statement contains a follow-up information point.

[0080] In some embodiments, the agent can construct a first prompt word based on the first reply sentence and the follow-up customization requirement. As an example only, the first prompt word can be "Please determine whether 'yyy1' contains a follow-up information point or whether it contains content with similar semantics. The follow-up information point can be 'error' or 'bug'." Among them, "yyy1" refers to the first reply sentence. The agent can input the first prompt word into the first large language model to obtain a judgment result. In some embodiments, when the first reply sentence contains a follow-up information point, step 342 is executed, otherwise step 343 is executed.

[0081] Step 342: Generate and output a first query statement.

[0082] In some embodiments, when the first reply statement contains a follow-up information point, the agent can construct a second prompt word "When 'yyy1' contains a follow-up information point, ask a follow-up question for the follow-up information point to obtain more explanations from the user on the follow-up information point", and the agent can input the second prompt word into the first large language model, so that the first large language model generates a first follow-up question for the follow-up information point. The first follow-up question is used to guide the target user to further describe the follow-up information point contained in the first reply statement.

[0083] In some embodiments, the first follow-up question may be displayed via a model dialogue interface.

[0084] Step 343, determining whether the first response statement completely responds to the first interview question.

[0085] In some embodiments, when the first reply statement does not contain a follow-up information point, the agent may construct a third prompt word "When 'yyy1' does not contain a follow-up information point, determine whether the user's reply completely replies to 'xxx1'; wherein the user's reply includes 'yyy1' and 'yyy2'", where "xxx1" refers to the first interview question, and "yyy2" refers to the historical sentences of the target user. The agent may input the third prompt word into the first large language model to obtain a corresponding judgment result. In some embodiments, when the first reply statement completely replies to the first interview question, step 344 may be executed, otherwise step 345 may be executed.

[0086] Step 344: Generate and output a query statement corresponding to the second interview question.

[0087] In some embodiments, the agent may construct a fourth prompt word "When the user's response fully answers 'xxx1', please ask the user a question based on 'xxx2'", where "xxx2" refers to the second interview question. The agent may input the fourth prompt word into the first large language model, and may output an inquiry sentence corresponding to the second interview question when the target user fully answers the first interview question. In some embodiments, the inquiry sentence may be output through a model dialogue interface for the target user to view.

[0088] Step 345: Generate and output a second query statement.

[0089] In some embodiments, the agent may construct a fifth prompt word "When the user's response does not fully answer 'xxx1', please continue to ask the user questions based on 'xxx1' and the user's response to get a more complete response". The agent may input the fifth prompt word into the first large language model to output a second follow-up statement when the target user does not fully respond to the first interview question. The second follow-up statement is used to guide the target user to fully respond to the first interview question. In some embodiments, the second follow-up statement may be output through the model dialogue interface for the target user to view.

[0090] In some embodiments, the first to fifth prompt words mentioned above may be input into the first large language model in steps, or may be combined together to form a longer prompt word and input into the first large language model at one time.

[0091] It should be understood that the aforementioned first interview question and the second interview question can be any two adjacent interview questions in the interview outline. When the inquiry question corresponding to the second interview question is displayed in the model dialogue interface, the current second interview question can be regarded as the first interview question, and steps 320 to 340 can be re-executed. In some embodiments, process 300 or part of the steps therein can be executed multiple times until all the interview questions in the interview outline are asked and the target user's reply statement is obtained.

[0092] In some embodiments, the intelligent agent can also call the first language model based on the interview outline to determine whether the interview questions in the interview outline have all been fully answered by the target user. When the interview questions in the interview outline have all been fully answered by the target user, an end sentence is generated and output. As an example, the intelligent agent can construct the prompt words "Please determine whether the interview questions in the interview outline have all been fully answered based on the user's reply. When they have all been fully answered, please output an end sentence; wherein the user's reply includes: 'yyy3', and the interview question includes 'xxx3'. "yyy3" may include all the reply sentences output by the target user in the interview dialogue, and "xxx3" may include all the interview questions in the interview outline. In some embodiments, the aforementioned end sentence can be displayed through the model dialogue interface to politely end the interview dialogue with the target user. In some embodiments, after determining that the target user has fully answered all the interview questions, the current interview task identifier can be added to the target user's historical interview identifier list.

[0093] In some embodiments, the agent can be implemented based on the third language model. Specifically, the third language model can be fine-tuned so that it can generate various prompt words for calling the first language model based on a preset prompt word template, thereby realizing the process management or control of the interview dialogue.

[0094] In some optional embodiments, the interview dialogue shown in step 230 can be implemented through voice dialogue. Specifically, the target user can reply to the voice for the interview. Similarly, the interrogation sentence and the follow-up question sentence output by the first language model can also be in voice form. The voice dialogue can be implemented on the basis of the text dialogue using text-to-speech TTS technology and speech-to-text STT technology. For example, the voice of the user's reply can be converted into a reply sentence in text form using STT technology. When the first language model generates an interrogation sentence and a follow-up question sentence in text form, it can be converted into voice using TTS technology and output to the model dialogue interface. The target user can click the play control to listen. Exemplarily, TTS technology may include coqui TTS, GPT-SoVITS, MeloTTS: High-qualit, etc., and STT technology may include coqui STT, whisper: Robust Speech R, PaddleSpeech: Easy, etc.

[0095] Step 240 , generating an interview report corresponding to the interview task based on the target user's reply statement. In some embodiments, step 240 may be implemented by a report generation model 540 .

[0096] After the interview dialogue is over, the target user's reply statements during the interview dialogue process are automatically stored. In some embodiments, the interview reply statements collected by the interview task can be classified together, prompt words can be constructed according to the interview questions and the reply statements under the questions, and the second language model can be used to summarize the target user's user views on the interview questions from the reply statements.

[0097] In some embodiments, the reply statements corresponding to each interview question of the interview task can be processed by the second largest language model to obtain one or more user opinions; and the interview report can be obtained based on the one or more user opinions corresponding to each interview question. In some embodiments, prompt words can be constructed based on the interview question and the reply statement of the target user under the interview question, and the second largest language model can be called to cluster these reply statements to obtain one or more reply statement clusters; and then one or more user opinions corresponding to the interview question are extracted based on the obtained reply statement clusters. As an example, each reply statement cluster under the interview question can be summarised and calculated as the corresponding user opinion of the cluster. More specifically, some reply statements can be selected from each reply statement cluster for summary calculation to obtain the corresponding user opinion. In other embodiments, one or more preset opinions pre-set by the second user for the interview question can also be obtained. The preset opinions can be keywords, short sentences, etc. that can reflect a certain opinion. The reply statements of the target user under a certain interview question are classified by the second largest language model and divided into one or more preset opinions corresponding to the interview question. As an example, the semantic similarity between the reply statement and each preset viewpoint can be calculated. When the maximum semantic similarity between the reply statement and the preset viewpoint is greater than a set threshold (such as 80%, 90%, etc.), the reply statement is divided into the preset viewpoint corresponding to the maximum semantic similarity. Finally, the preset viewpoint with the reply statement is used as the aforementioned user viewpoint. For example, the second user pre-sets preset viewpoint 1, preset viewpoint 2, and preset viewpoint 3 for interview question 1. For interview task A, by executing steps 210 to 230, 1,000 reply statements from different target users for interview question 1 are obtained. These 1,000 reply statements are divided into preset viewpoint 1 and preset viewpoint 3 according to semantic similarity, and the preset viewpoint 2 is not divided to obtain a reply statement. Therefore, preset viewpoint 1 and preset viewpoint 3 can be used as user views corresponding to interview question 1, respectively. In some embodiments, the second largest language model and the first largest language model can be the same large language model, or can be another large language model different from the first large language model.

[0098] In some embodiments, the interview report may include an interview task identifier, interview questions recorded in the interview outline, and user opinions corresponding to each interview question. In other embodiments, the interview report may also list a number of corresponding reply statements under each user opinion so that the second user can understand the target user's direct reply to the relevant interview questions.

[0099] In some optional embodiments, the interview report may also include information such as the number of interview task completions (e.g., each time a target user completes an interview task, the number is incremented by 1), the average interview duration (e.g., the average duration of interviews for different target users to complete the interview task), etc. The second user may quickly analyze the first user's satisfaction with the product, the responses to the interview questions, and the overall interview quality in this interview survey based on the interview report.

[0100] Figure 5 is an exemplary frame of an automated interview system according to some embodiments of this specification. Figure 5 As shown, the automated interview system 500 provided in some embodiments of the present specification may include a configuration module 510 , a distribution module 520 , a dialogue module 530 , and a report generation module 540 .

[0101] The configuration module 510 is used to obtain interview task configuration information; wherein, the interview task configuration information includes interviewee selection criteria and an interview outline including one or more interview questions.

[0102] The distribution module 520 is used to determine the target user to be interviewed based on the interviewee selection conditions and send an interview request to the target user.

[0103] The dialogue module 530 is used for conducting one or more dialogues with the target user based on the interview outline through the first language model in response to the target user accepting the interview request, so as to obtain a reply statement from the target user.

[0104] The report generation module 540 is used to generate an interview report corresponding to the interview task based on the reply statement of the target user.

[0105] For more information about each module, see Figure 2~Figure 4 The relevant description of will not be repeated here. It should be understood that Figure 5The system and its modules shown can be implemented in various ways. For example, in some embodiments, the system and its modules can be implemented by hardware, software, or a combination of software and hardware. Among them, the hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. Those skilled in the art will understand that the above methods and systems can be implemented using computer executable instructions and / or control codes contained in a processor, such as a disk, CD or DVD-ROM, etc., and such codes are provided in a carrier medium or a memory of a programmable device. The system and its modules of this specification can not only be implemented by hardware circuits such as ultra-large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., but can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above hardware circuits and software (e.g., firmware).

[0106] It should be noted that the above description of the system and its modules is only for convenience of description and does not limit this specification to the scope of the embodiments. It is understandable that for those skilled in the art, after understanding the principle of the system, it is possible to arbitrarily combine the modules without deviating from this principle to form a subsystem connected to other modules. Or some modules may be split to obtain more modules or multiple units under the module. Such variations are all within the scope of the disclosure of this specification.

[0107] Some embodiments of this specification also provide a computer program product, including computer instructions or computer codes. When at least part of the computer instructions or computer codes are executed by a processor, the automated interview method shown in some embodiments of this specification can be implemented. In some embodiments, the computer program product may only involve computer instructions or computer codes, which may be carried by a storage medium or a processing device. In other embodiments, the computer program product may also be a storage medium or a processing device containing the aforementioned computer instructions or computer codes. The processing device may include one or more processors, and a storage medium.

[0108] In some embodiments, the processor may be a combination of one or more of the following processors: a central processing unit (CPU), an application specific integrated circuit (ASIC), an application specific instruction set processor (ASIP), a graphics processing unit (GPU), a physical processing unit (PPU), a digital signal processor (DSP), a field programmable gate array (FPGA), a programmable logic device (PLD), a programmable logic controller (PLC), a reduced instruction set computer (RISC), and a microprocessor.

[0109] In some embodiments, the storage medium may include a combination of one or more of the following: a mass storage, a removable storage, a volatile read-write memory, and a read-only memory (ROM). Exemplary mass storage may include a magnetic disk, an optical disk, a solid-state hard disk, and the like. Exemplary removable storage may include a flash disk, a floppy disk, an optical disk, a memory card, a compressed hard disk, a magnetic tape, and the like. Exemplary volatile read-write memory may include a random access memory (RAM). Exemplary random access memory may include a dynamic random access memory (DRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), a static random access memory (SRAM), a thyristor random access memory (T-RAM), and a zero capacitance memory (Z-RAM), and the like. Exemplary read-only memory may include a masked read-only memory (MROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a compact disk read-only memory (CD-ROM), and a digital multifunction hard disk read-only memory, and the like.

[0110] The beneficial effects that may be brought about by the embodiments of this specification include but are not limited to: (1) Compared with manual interviews, the automated interview method provided by the embodiments of this specification can significantly reduce labor costs and research cycles; (2) It can distribute interview requests to precise timing and precise groups based on the product usage trends and user portraits of the first user; (3) It can lead or manage the dialogue process based on the interview outline, customized requirements for follow-up questions, and intelligent user responses, so that the interview dialogue can be carried out efficiently around the interview task; (4) It can autonomously distinguish the first user's response and autonomously ask questions about incomplete responses or points of interest in the response, thereby ensuring the quality of the interview dialogue; (5) After the interview dialogue is over, the interview dialogue can be automatically analyzed to generate an interview report, so that the second user can quickly understand the user's views and the interview situation. It should be noted that different embodiments may produce different beneficial effects. In different embodiments, the beneficial effects that may be produced may be any one or a combination of the above, or any other beneficial effects that may be obtained.

[0111] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements and corrections to this specification. Such modifications, improvements and corrections are taught in this specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of this specification.

Claims

1. An automated interview method, characterized in that: include: Acquire interview task configuration information; wherein the interview task configuration information includes interviewee selection criteria and an interview outline including one or more interview questions; Based on the interviewee selection criteria, determine the target user to be interviewed and send an interview request to the target user; In response to the target user accepting the interview request, conducting one or more rounds of dialogue with the target user based on the interview outline through the first language model to obtain a reply statement from the target user; An interview report corresponding to the interview task is generated based on the reply statement of the target user.

2. The method according to claim 1, characterized in that Determining the target user to be interviewed based on the interviewee selection condition includes: Determine a user status query condition and / or a user behavior query condition based on the interviewee selection condition; Determine a first user who meets the user status query condition based on a first database; wherein the first database is used to maintain basic personal information and / or historical behavior statistics of one or more first users; Determine a first user who meets the user behavior query condition based on a second database; wherein the second database is used to maintain behavior data of one or more first users; The target user to be interviewed is determined based on the first user who meets the user status query condition and / or the first user who meets the user behavior query condition.

3. The method according to claim 2, characterized in that The first user includes a game player; The first user's basic personal information includes: account identification, game character information, and game level; The historical behavior statistics of the first user include one or more of the following information: login time, game time, number of games, and consumption status; The behavior data of the first user comes from a behavior log of the first user in the game.

4. The method according to claim 1, characterized in that The interview task configuration information also includes interview time conditions; The sending of the interview request to the target user further includes: In response to the current moment satisfying the interview time condition, sending the interview request to the user terminal corresponding to the target user.

5. The method according to claim 1, characterized in that The sending of the interview request to the target user also includes: obtaining a historical interview identifier list of the target user, the historical interview identifier list being used to indicate historical interview tasks in which the target user has participated; when it is determined based on the historical interview identifier list that the target user has not participated in the current interview task, sending the interview request to the target user; The method further includes: after obtaining the reply statement of the target user, updating the historical interview identification list of the target user based on the current interview task.

6. The method according to claim 1, characterized in that The determining of the target user to be interviewed based on the interviewee selection condition and sending an interview request to the target user comprises: Determining an interview triggering behavior based on the interviewee selection criteria; When there is behavior data of the first user matching the interview triggering behavior in the third database, the first user corresponding to the behavior data is determined as the target user and an interview request is sent to the first user; wherein the third database is used to obtain and store one or more behavior logs of the first user in the preset application system in real time.

7. The method according to claim 1, characterized in that The interview task configuration information also includes interview purpose, interview background knowledge and conversation style information, and the interview request includes a link to a model conversation interface for conducting a conversation based on a model; In response to the target user accepting the interview request, conducting one or more rounds of dialogue with the target user based on the interview outline through the first language model to obtain a reply statement of the target user, further comprising: In response to the target user entering the model dialogue interface through the link, generating an interview opening statement through a first large language model based on at least one of the following information: the purpose of the interview, the background knowledge of the interview, and the dialogue style information; The interview opening statement is displayed on the model dialogue interface.

8. The method according to claim 1, characterized in that The conducting one or more rounds of dialogue with the target user based on the interview outline by using the first language model includes: The first large language model is called by an intelligent agent based on the first interview question in the interview outline to generate and output a query sentence corresponding to the first interview question; In response to receiving the first reply statement of the target user, the agent calls the first large language model to determine whether the target user has an ending intention based on the first reply statement or based on the first reply statement and the historical statements of the target user; In response to the judgment result that the target user has an ending intention, the agent calls the first language model to generate and output an ending sentence; In response to the judgment result that the target user has no intention to end the interview, the intelligent agent calls the first large language model to perform follow-up questions or raise the next interview question in the interview outline.

9. The method according to claim 8, characterized in that The interview task configuration information also includes follow-up customized requirements; The step of calling the first large language model to perform follow-up questions or raise the next interview question in response to the judgment result that the target user does not have the intention to end the interview by the intelligent agent includes: The agent calls the first large language model to generate and output a follow-up question sentence or an inquiry sentence corresponding to the second interview question based on the first interview question, the first reply sentence, the second interview question in the interview outline, and a follow-up question customization requirement, or based on the first interview question, the first reply sentence, the second interview question in the interview outline, a follow-up question customization requirement, and the target user's historical sentences; The follow-up statement is used to guide the target user to completely reply to the first interview question and / or to guide the target user to further describe the information points contained in the first reply statement.

10. The method according to claim 9, characterized in that The customized inquiry requirement further includes inquiry information points; The agent calls the first large language model to generate and output a follow-up question or an inquiry sentence corresponding to the second interview question based on the first interview question, the first reply sentence, the second interview question in the interview outline, and a follow-up customized requirement, or based on the first interview question, the first reply sentence, the second interview question in the interview outline, a follow-up customized requirement, and the target user's historical sentence, including calling the first large language model to perform the following processing based on the first interview question, the first reply sentence, the second interview question in the interview outline, and a follow-up customized requirement, or based on the first interview question, the first reply sentence, the second interview question in the interview outline, a follow-up customized requirement, and the target user's historical sentence: Determining whether the first reply statement contains the follow-up information point; When the first reply statement contains the follow-up information point, a first follow-up question statement is generated and output; wherein the first follow-up question statement is used to guide the target user to further describe the follow-up information point contained in the first reply statement; When the first reply statement does not include the follow-up information point, determining whether the first reply statement completely answers the first interview question; When the first reply statement completely answers the first interview question, generating and outputting an inquiry statement corresponding to the second interview question; When the first reply statement does not fully answer the first interview question, a second follow-up question statement is generated and output; wherein the second follow-up question statement is used to guide the target user to fully answer the first interview question.

11. The method according to claim 8, characterized in that In response to the target user accepting the interview request, conducting one or more rounds of dialogue with the target user based on the interview outline through the first language model, further comprising: The intelligent agent calls the first language model based on the interview outline to determine whether all the interview questions in the interview outline have been completely answered by the target user. When all the interview questions in the interview outline have been completely answered by the target user, an end statement is generated and output.

12. The method according to claim 1, characterized in that The generating of an interview report corresponding to the interview task based on the reply statement of the target user includes: Using the second language model to process the response sentences corresponding to the interview questions of the interview task, and obtain one or more user opinions; The interview report is obtained based on one or more user viewpoints corresponding to each interview question.

13. The method according to claim 12, characterized in that The second language model is used to process the response sentences corresponding to the interview questions of the interview task respectively to obtain one or more user opinions, including: Clustering the response sentences corresponding to the same interview question respectively by using the second largest language model to obtain one or more response sentence clusters; extracting one or more user opinions corresponding to the interview question based on the one or more response sentence clusters; Alternatively, based on one or more preset viewpoints corresponding to the same interview question, the second largest language model divides the reply sentences corresponding to the interview question into the one or more preset viewpoints; and obtains one or more user viewpoints corresponding to the interview question based on the preset viewpoints corresponding to the reply sentences.

14. The method according to claim 1, characterized in that The first large language model corresponds to an interview task; wherein the first large language model is fine-tuned based on the interview background knowledge and / or dialogue style information of the corresponding interview task.

15. An automated interview system, characterized in that: include: A configuration module, used to obtain interview task configuration information; wherein the interview task configuration information includes interviewee selection conditions and an interview outline including one or more interview questions; A distribution module, used to determine the target user to be interviewed based on the interviewee selection conditions and send an interview request to the target user; A dialogue module, configured to, in response to the target user accepting the interview request, conduct one or more rounds of dialogue with the target user based on the interview outline through a first language model to obtain a reply statement from the target user; A report generation module is used to generate an interview report corresponding to the interview task based on the reply statement of the target user.

16. A computer program product, comprising computer codes, which, when at least part of the computer codes are executed by a processor, can implement the method according to any one of claims 1 to 14.

Images