Question recommendation method and device, storage medium and electronic equipment

Abstract

The invention discloses a question recommendation method and device, a storage medium and electronic equipment, and the method comprises the steps: determining a plurality of historical questions proposed by a user in a current session, and employing a pre-trained question prediction model according to each historical question, and determining a prediction question outputted by the question prediction model. Determining a first question matched with the predicted question from a plurality of first questions included in a pre-constructed question and answer database, and taking the first question as a to-be-selected question; and according to the to-be-selected questions, determining a target question, and recommending the target question to the user. The historical problem in the current session is captured, the prediction problem conforming to the current session context is determined based on the historical problem by adopting the large language model, the to-be-selected problem matched with the prediction problem is determined, and the target problem recommended to the user is obtained based on the to-be-selected problem, so that the accuracy of the target problem is improved, and the user experience is improved. And the target question recommended to the user better meets the user demand, so that the user experience is improved.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and particularly to a method, apparatus, storage medium, and electronic device for problem recommendation. Background Art

[0002] With the continuous development of technology, intelligent customer service has received extensive attention and has been widely applied in various industries.

[0003] Currently, when a user has a problem during the process of business execution, the user can ask the intelligent customer service a question. The intelligent customer service can promptly respond to the question raised by the user, determine the answer to the question, and return it to the user. At the same time, the intelligent customer service can also automatically recommend questions to the user so that the user can directly select the recommended questions, and the intelligent customer service directly returns the answers to the questions selected by the user to the user without the user having to manually input the questions, improving the user experience. Therefore, how to accurately recommend questions is a very important issue.

[0004] Based on this, this specification provides a method for problem recommendation. Summary of the Invention

[0005] This specification provides a method, apparatus, storage medium, and electronic device for problem recommendation to partially solve the above problems existing in the prior art.

[0006] This specification adopts the following technical solutions:

[0007] This specification provides a method for problem recommendation, including:

[0008] Determine a number of historical questions raised by the user in the current session;

[0009] According to each historical question, use a pre-trained question prediction model to determine the predicted question output by the question prediction model; wherein, the question prediction model is a large language model;

[0010] From a number of first questions included in a pre-constructed Q&A database, determine the first question that matches the predicted question and use it as the candidate question;

[0011] According to the candidate question, determine the target question and recommend the target question to the user.

[0012] Optionally, according to each historical question, using a pre-trained question prediction model to determine the predicted question specifically includes:

[0013] Determine the key question from each historical question;

[0014] Input the key problem into a pre-trained problem prediction model to determine the predicted problem output by the problem prediction model.

[0015] Optionally, determine the key problem from each historical problem, specifically including:

[0016] Screen each historical problem to determine the historical problems within a preset time range, and use them as the second problems;

[0017] Input the second problems into the problem prediction model to determine the key problems output by the problem prediction model.

[0018] Optionally, determine the first problem that matches the predicted problem from several first problems included in a pre-constructed problem database, and use it as the candidate problem, specifically including:

[0019] For each first problem included in the pre-constructed Q&A database, determine the first similarity between this first problem and the predicted problem;

[0020] According to the first similarities respectively corresponding to each first problem, sort the first problems in descending order of the first similarity to obtain a problem sequence;

[0021] Determine a specified number of first problems in the order of the problem sequence, and use them as the candidate problems that match the predicted problem.

[0022] Optionally, determine the target problem according to the candidate problems, specifically including:

[0023] For each candidate problem, determine the second similarities respectively corresponding between this candidate problem and each historical problem;

[0024] Judge whether there is a second similarity that reaches a preset threshold among the second similarities corresponding to this candidate problem;

[0025] If so, do not use this candidate problem as the target problem;

[0026] If not, use this candidate problem as the target problem.

[0027] Optionally, pre-train the problem prediction model, specifically including:

[0028] Obtain the problem data in the historical session;

[0029] Use the problems within the first time period in the problem data as the first samples, and use the problems within the second time period in the problem data as the first annotations of the first samples; wherein, the first time period is earlier than the second time period;

[0030] Input the first sample into the problem prediction model to be trained, and determine the prediction result output by the problem prediction model to be trained;

[0031] Train the problem prediction model to be trained according to the prediction result and the first annotation.

[0032] Optionally, the method further includes:

[0033] In response to the click operation of the user, determine the target problem selected by the user and use it as the question to be replied;

[0034] Determine the first reply corresponding to the question to be replied from several first replies included in the Q&A database and use it as the target reply;

[0035] Display the target reply to the user.

[0036] This specification provides a device for question recommendation, including:

[0037] A determination module, configured to determine several historical questions proposed by the user in the current session;

[0038] A prediction module, configured to determine the predicted question output by the problem prediction model according to each historical question by using a pre-trained problem prediction model; wherein, the problem prediction model is a large language model;

[0039] A screening module, configured to determine a first question that matches the predicted question from several first questions included in a pre-constructed Q&A database and use it as a candidate question;

[0040] A recommendation module, configured to determine a target question according to the candidate question and recommend the target question to the user.

[0041] This specification provides a computer-readable storage medium, where the storage medium stores a computer program, and when the computer program is executed by a processor, the method for question recommendation described above is implemented.

[0042] This specification provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the method for question recommendation described above is implemented.

[0043] At least one of the technical solutions adopted in this specification can achieve the following beneficial effects:

[0044] The method for question recommendation provided in this specification determines several historical questions raised by the user in the current session. Based on each historical question, a pre-trained question prediction model is used to determine the predicted question output by the question prediction model. Then, from several first questions included in the pre-constructed Q&A database, a first question that matches the predicted question is determined and used as a candidate question. Then, based on the candidate question, a target question is determined and recommended to the user.

[0045] As can be seen from the above method, when recommending questions in this application, several historical questions raised by the user in the current session are determined. Based on each historical question, a pre-trained question prediction model is used to determine the predicted question output by the question prediction model. Then, from several first questions included in the pre-constructed Q&A database, a first question that matches the predicted question is determined and used as a candidate question. Then, based on the candidate question, a target question is determined and recommended to the user. By capturing the historical questions in the current session and then based on the historical questions, using a large language model to determine the predicted questions that conform to the current conversation context and determining the candidate questions that match the predicted questions, so as to obtain the target questions recommended to the user based on the candidate questions, the accuracy of the target questions is improved, and the target questions recommended to the user are more in line with the user's needs, thereby improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings described herein are used to provide a further understanding of this specification and form a part of this specification. The schematic embodiments of this specification and their descriptions are used to explain this specification and do not constitute an improper limitation to this specification. In the drawings:

[0047] Figure 1 It is a schematic flow chart of a method for question recommendation provided in this specification;

[0048] Figure 2 It is a schematic diagram of a device for question recommendation provided in this specification;

[0049] Figure 3 It corresponds to a Figure 1 structural schematic diagram of an electronic device. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] To make the purpose, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this specification.

[0051] Currently, when a user has a problem with the process of business execution, the user can start a conversation with the intelligent customer service and ask the intelligent customer service a question. The intelligent customer service can respond to the question raised by the user in a timely manner, determine the answer to the question, and return it to the user. At the same time, in the current conversation, the intelligent customer service can also automatically recommend questions to the user so that the user can directly select the recommended questions, and the intelligent customer service directly returns the answers to the questions selected by the user. When recommending questions, it mainly relies on the user's historical data and user profile to determine the questions to be recommended to the user and make recommendations. However, this method still has deficiencies in terms of real-time performance and accuracy, and over-reliance on historical data and user profiles may also lead to information redundancy and a decline in user experience.

[0052] Based on this, this specification provides a method, device, storage medium, and electronic device for question recommendation. The following will combine with the accompanying drawings to detail the technical solutions provided by each embodiment of this specification.

[0053] Figure 1 The flowchart of a method for question recommendation provided in this specification includes the following steps:

[0054] S100: Determine several historical questions raised by the user in the current conversation.

[0055] In this specification, the device for question recommendation can first determine several historical questions raised by the user in the current conversation. Among them, the device for question recommendation can be an intelligent customer service, the server or system of the intelligent customer service, or an electronic device such as a desktop computer or a laptop. For the convenience of description, hereinafter, only the server is used as the execution subject to illustrate the method for question recommendation provided in this specification. The above-mentioned current conversation is the conversation currently in progress between the current user and the intelligent customer service. The user and the intelligent customer service can have multiple conversations, that is, the user can start multiple conversations with the intelligent customer service. When the user asks the intelligent customer service a question for the first time or after a long time, it means that the user starts a conversation with the intelligent customer service. In the conversation, the user and the intelligent customer service conduct multiple rounds of question and answer, that is, the user can ask the intelligent customer service questions multiple times, and the intelligent customer service can return the answers to the questions raised by the user multiple times. In the conversation, if the user does not ask the intelligent customer service a question for a long time, the conversation is defaulted to end. When the user asks the intelligent customer service a question again after the conversation ends, that is, asks the intelligent customer service a question again after a long time, at this time, it means that the user starts a conversation with the intelligent customer service again.

[0056] The above historical question is a question that the user has already asked the intelligent customer service in the current session, that is, the historical question is a question that the user asked the intelligent customer service before the current moment in the current session. The historical question can be the question entered by the user. The historical question can be a problem encountered by the user during the execution of the business, and the intelligent customer service can be the customer service corresponding to the business, and the intelligent customer service is used to answer the questions raised by the user. The above business can be a game, the above user can be a game player, and the above intelligent customer service can be a game customer service. Of course, the above business can also be payment, and the above intelligent customer service can also be a payment customer service, which is not specifically limited in this specification. For the convenience of description, the following business takes a game as an example, and the user takes a game player as an example. Based on this, the above historical question can be a question related to the game, such as a question related to the game operation process, or a question related to the recharge in the game.

[0057] S102: According to each historical question, use a pre-trained question prediction model to determine the predicted question output by the question prediction model; wherein, the question prediction model is a large language model.

[0058] In this specification, the server can use a pre-trained question prediction model according to each historical question to determine the predicted question output by the question prediction model. Among them, the above question prediction model is a pre-trained model, and the question prediction model is a large language model. The question prediction model can be used to predict the potential questions that the user may ask in the next round according to each historical question. Specifically, the server can directly input each historical question into the pre-trained question prediction model to determine the predicted question output by the question prediction model. The predicted question is the potential question that the user may ask in the next round.

[0059] In this specification, the above historical question can be all the questions asked by the user in the history of the current session. Of course, the above historical question can also be some of the questions asked by the user in the history of the current session. In order to make the questions predicted by the question prediction model more in line with the context of the current session, the server can first screen the historical questions, and then determine the predicted questions based on the screened historical questions. Based on this, the server can screen each historical question to determine the historical questions within a preset time range and use them as the second questions. Then input the second questions into the pre-trained question prediction model to determine the predicted questions output by the question prediction model. Among them, the preset time range can be a preset time range, and the preset time range is a historical time period, that is, the historical time period before the current moment. In order to ensure that the predicted questions are more in line with the questions that the user may ask in the next round, the preset time range is the time period closest to the current moment. Then the preset time range can be the past ten minutes, or it can be the past five minutes, which is not specifically limited in this specification.

[0060] In addition, to ensure that the questions of the input question prediction model are the key questions in the current conversation and are more in line with the context of the current conversation, the server can also determine the key questions from each historical question. Then, the key questions are input into a pre-trained question prediction model to determine the predicted questions output by the question prediction model. Among them, the above key questions can be at least one of the historical questions. When determining the key questions from each historical question, the server can use a pre-set screening algorithm or a pre-trained screening model to screen each historical question and use the screened historical questions as the key questions. The screening algorithm can be a pre-set algorithm or any existing screening algorithm, which is not specifically limited in this specification. The screening algorithm is used to screen historical questions to determine the key questions. The screening model can be a pre-trained machine learning model or any existing screening model, which is not specifically limited in this specification. The screening model is used to screen historical questions to determine the key questions.

[0061] In addition, the above screening model can also be the above question prediction model, that is, the question prediction model can also be used to screen each historical question to determine the key questions from each historical question. Therefore, when determining the key questions from each historical question, the server can also input each historical question into the question prediction model to screen each historical question through the question prediction model and determine the key questions.

[0062] Furthermore, the server can first screen each historical question to determine the historical questions within a preset time range and use them as the second questions, and then input the second questions into the question prediction model to determine the key questions output by the question prediction model, so that the predicted questions output by the question prediction model are more in line with the context of the current conversation.

[0063] S104: Determine a first question that matches the predicted question from several first questions included in the pre-constructed Q&A database and use it as the candidate question.

[0064] In this specification, the server can determine, from several first questions included in a pre-constructed Q&A database, a first question that matches the predicted question and use it as a candidate question. The Q&A database is a pre-constructed database that includes several first questions and several first answers. Each first question has a corresponding first answer, that is, there is a one-to-one correspondence between the first questions and the first answers. Each of the above first questions is an existing question. These first questions can be questions that all users of the game have asked in history, or questions preset by game operators. This specification does not make specific limitations. And each first answer can be an answer returned by an intelligent customer service based on each first question (that is, questions that all users of the game have asked in history), or an answer preset by game operators based on each first question (that is, questions preset by game operators). This specification does not make specific limitations. Of course, each first question can also be questions that all users of the game have asked in history and questions preset by game operators. Correspondingly, each first answer can also be an answer returned by an intelligent customer service based on the questions asked by users and an answer set by game operators based on the preset questions.

[0065] Specifically, the server can determine each first question included in the pre-constructed Q&A database, and then input each first question and the predicted question into a matching model to determine, through the matching model, a first question that matches the predicted question from each first question and use it as a candidate question. Among them, the above matching model can be a pre-trained model or any existing model. The matching model is used to determine a question that matches the predicted question from each first question. The above candidate question is at least one first question in the Q&A database, that is, at least one first question in the Q&A database matches the predicted question.

[0066] In addition, the server can also determine a first similarity between each first question included in the pre-constructed Q&A database and the predicted question. According to the first similarities corresponding to each first question, determine, from each first question, a first question that matches the predicted question and use it as a candidate question. Among them, the above first similarity characterizes the similarity between the first question and the predicted question. When determining the first similarity between the first question and the predicted question, the server can use any existing similarity algorithm or model to determine the first similarity between the first question and the predicted question. This specification does not make specific limitations.

[0067] When determining, based on the first similarity corresponding to each first question, a first question that matches the predicted question from each first question and using it as a candidate question, the server may also, for each first question, determine whether the first similarity corresponding to the first question reaches a specified threshold. If so, the first question is used as a candidate question that matches the predicted question; if not, the first question is not used as a candidate question that matches the predicted question. Herein, the specified threshold may be a preset value, which may be 0.9 or may also be 0.8. This specification does not make specific limitations.

[0068] In addition, the server may also sort each first question according to the first similarity corresponding to each first question in descending order of the first similarity to obtain a question sequence. Then, according to the order of the question sequence, a specified number of first questions are determined and used as candidate questions that match the predicted question. Herein, the above-specified number may be a preset value, which may be 1 or may also be 5. This specification does not make specific limitations. The above candidate questions are first questions in the Q&A database, and there is at least one such candidate question.

[0069] S106: Determine a target question according to the candidate question and recommend the target question to the user.

[0070] In this specification, the server may determine a target question according to the candidate question and recommend the target question to the user. Specifically, the server may directly use the candidate question as the target question and recommend the target question to the user. When recommending the target question to the user, the server may directly display the target question to the user. Of course, the server may also generate target options corresponding to the target question and then display the target options to the user.

[0071] In this specification, in order to ensure that the target question is not a question previously asked by the user and effectively avoid repeated recommendations and information redundancy, the server may determine a second similarity between each historical question and the candidate question for each historical question. Then, it is determined whether there is a second similarity that reaches a preset threshold among the second similarities, that is, it is determined whether there is at least one second similarity that reaches the preset threshold among the second similarities. If so, the candidate question is not used as the target question; if not, the candidate question is used as the target question.

[0072] Among them, the preset threshold can be a preset value, which can be 0.9 or 0.8. This specification does not make specific limitations. The second similarity characterizes the similarity degree between the candidate question and the historical questions. When there is a second similarity among the second similarities that reaches the preset threshold, it indicates that there is at least one second similarity among the second similarities that reaches the preset threshold, and further indicates that there is at least one historical question with a high similarity to the candidate question. At this time, the candidate question is not used as the target question. However, when there is no second similarity among the second similarities that reaches the preset threshold, it means that none of the second similarities reaches the preset threshold, and further characterizes that each historical question has a low similarity to the candidate question. At this time, the candidate question is used as the target question.

[0073] In this specification, there can be one candidate question. Therefore, when there is only one candidate question and there is a second similarity among the second similarities that reaches the preset threshold, the candidate question is not used as the target question. At this time, the server can re-determine the candidate question, that is, re-execute the above-mentioned step S104 to re-determine the candidate question. Of course, the server can also generate a first prompt message and display it to the game operator to determine the candidate question by the game operator.

[0074] However, there can also be multiple candidate questions as mentioned above. Therefore, when determining the target question based on the candidate questions, the server can determine the second similarities corresponding to each candidate question and each historical question respectively for each candidate question, and then determine whether there is a second similarity that reaches the preset threshold among the second similarities corresponding to the candidate question. If so, the candidate question is not used as the target question. If not, the candidate question is used as the target question. Among them, the process of determining the second similarities corresponding to each candidate question and each historical question respectively is similar to the process of determining the second similarities mentioned above, and will not be elaborated here.

[0075] It can be seen from the above method that when making question recommendations, the server can determine several historical questions proposed by the user in the current session, and according to each historical question, use a pre-trained question prediction model to determine the predicted question output by the question prediction model. Then, from several first questions included in the pre-constructed Q&A database, determine the first question that matches the predicted question and use it as the candidate question. Then, according to the candidate question, determine the target question and recommend the target question to the user. By capturing the historical questions in the current session, and then based on the historical questions, using a large language model to determine the predicted question that conforms to the current conversation context, and determining the candidate question that matches the predicted question, so as to obtain the target question recommended to the user based on the candidate question, the accuracy of the target question is improved, and the target question recommended to the user is more in line with the user's needs, thus improving the user experience.

[0076] In this specification, the server can determine several historical questions raised by the user in the current session in real time, and recommend questions to the user according to the processes of steps S102 to S106 above. Of course, the server can also start to determine several historical questions raised by the user in the current session when the user does not raise questions for a long time, that is, when the user does not raise questions for a long time in the current session, and recommend questions to the user according to the processes of steps S102 to S106 above. Therefore, in step S100 above, when the waiting time in the current session reaches the preset duration, several historical questions raised by the user in the current session are determined. The waiting time is the duration from the time when the user last raised a question in the current session to the current moment, and the preset duration can be a preset value.

[0077] In this specification, there can also be multiple above-mentioned predicted questions. For each predicted question, the candidate questions matching each predicted question can be determined according to the process of step S104 above, that is, there can be multiple above-mentioned candidate questions, that is, each predicted question can have at least one candidate question matching it.

[0078] In this specification, after recommending the target question to the user, the server can also determine the target question selected by the user in response to the user's click operation, and use it as the question to be replied. Then, from several first replies included in the Q&A database, the first reply corresponding to the question to be replied is determined and used as the target reply. Then, the target reply is displayed to the user.

[0079] In this specification, the above-mentioned question prediction model can be a pre-trained large language model. Therefore, when pre-training the question prediction model, the server can obtain the question data in the historical session. The questions in the first time period in the question data are used as the first samples, and the questions in the second time period in the question data are used as the first annotations of the first samples. Then, the first samples are input into the question prediction model to be trained, and the prediction results output by the question prediction model to be trained are determined. According to the prediction results and the first annotations, the question prediction model to be trained is trained.

[0080] Among them, the first time period is earlier than the second time period, that is, the first time period precedes the second time period. The above-mentioned question data includes several questions. When training the question prediction model to be trained according to the prediction results and the first annotations, the server can determine the first loss according to the prediction results and the first annotations, and then train the question prediction model to be trained according to the first loss. The trained question prediction model can be used to determine the potential questions that may be raised in the next round of the input question, that is, the output questions.

[0081] In addition to predicting potential questions that may be raised in the next round, the above-mentioned question prediction model can also determine key questions from a number of questions. Therefore, when training the question prediction model to be trained, the server can also use a number of questions in the question data as the second sample, and use the key questions in each question as the second annotation of the second sample. Then, the second sample is input into the question prediction model to be trained to determine the recognition result, and then the question prediction model to be trained is trained according to the recognition result and the second annotation.

[0082] Among them, the above-mentioned second annotation can be determined by game operators from each question included in the second sample. The above-mentioned recognition result can be a question among a number of questions included in the second sample, that is, a key question. When training the question prediction model to be trained according to the recognition result and the second annotation, the server can also determine the second loss according to the recognition result and the second annotation, and then train the question prediction model to be trained according to the second loss, so that the trained question prediction model can also determine key questions from a number of input questions.

[0083] In this specification, the question prediction model to be trained can be a large language model that has not been trained, that is, the model parameters of the question prediction model to be trained are all initialized parameters, and the initialized parameters can be randomly initialized parameters. Of course, the question prediction model to be trained can also be a trained large language model, that is, any existing large language model, which is not specifically limited in this specification.

[0084] The above is the method of one or more implementations of this specification. Based on the same idea, this specification also provides a corresponding question recommendation device, such as Figure 2 shown.

[0085] Figure 2 The figure is a schematic diagram of a question recommendation device provided by this specification, including:

[0086] A determination module 200, configured to determine a number of historical questions raised by the user in the current session;

[0087] A prediction module 202, configured to determine a predicted question output by the question prediction model according to each historical question by using a pre-trained question prediction model; wherein, the question prediction model is a large language model;

[0088] A screening module 204, configured to determine a first question that matches the predicted question from a number of first questions included in a pre-constructed Q&A database and use it as a candidate question;

[0089] A recommendation module 206, configured to determine a target question according to the candidate question and recommend the target question to the user.

[0090] Optionally, the prediction module 202 is specifically configured to determine key questions from each historical question; input the key questions into a pre-trained question prediction model to determine the predicted questions output by the question prediction model.

[0091] Optionally, the prediction module 202 is specifically configured to screen each historical question to determine historical questions within a preset time range as the second questions; input the second questions into the question prediction model to determine the key questions output by the question prediction model.

[0092] Optionally, the screening module 204 is specifically configured to, for each first question included in the pre-constructed Q&A database, determine the first similarity between the first question and the predicted question; sort the first questions according to the first similarities corresponding to the first questions in descending order of the first similarities to obtain a question sequence; determine a specified number of first questions in the order of the question sequence as candidate questions matching the predicted question.

[0093] Optionally, the recommendation module 206 is specifically configured to, for each candidate question, determine the second similarities respectively corresponding between the candidate question and each historical question; determine whether there is a second similarity reaching a preset threshold among the second similarities corresponding to the candidate question; if so, not use the candidate question as the target question; if not, use the candidate question as the target question.

[0094] Optionally, the device further includes:

[0095] A training module 210, configured to obtain question data in a historical session; use the questions within a first time period in the question data as the first samples, and use the questions within a second time period in the question data as the first annotations of the first samples, where the first time period is earlier than the second time period; input the first samples into a question prediction model to be trained to determine the prediction results output by the question prediction model to be trained; train the question prediction model to be trained according to the prediction results and the first annotations.

[0096] Optionally, the device further includes:

[0097] A reply module 208, configured to, in response to the user's click operation, determine the target question selected by the user as the question to be replied; determine the first reply corresponding to the question to be replied from several first replies included in the Q&A database as the target reply; display the target reply to the user.

[0098] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the aboveFigure 1 A method for problem recommendation is provided.

[0099] This manual also provides Figure 3 The one shown corresponds to Figure 1 A schematic diagram of the electronic device. Figure 3 As shown, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Recommended approach to the problem described.

[0100] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0101] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many method flow improvements today can be regarded as direct improvements to hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structures by programming the improved method flows into the hardware circuits. Therefore, it cannot be said that an improvement to a method flow cannot be implemented using a hardware entity module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by a user's programming of the device. The designer can program it himself to "integrate" a digital system onto a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL). There is not just one type of HDL, but many types, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones currently are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that by simply performing a little logical programming on the method flow using the above-mentioned several hardware description languages and programming it into an integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0102] The controller may be implemented in any suitable manner. For example, the controller may take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program the method steps to enable the controller to be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same function. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0103] The systems, devices, modules, or units illustrated in the above embodiments may be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0104] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing this specification, the functions of each unit may be implemented in the same or multiple software and / or hardware.

[0105] Those skilled in the art should understand that the embodiments of this specification may be provided as a method, a system, or a computer program product. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0106] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It should be understood that each flow and / or block in the flowchart and / or block diagram, and combinations of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions executed by the processor of the computer or other programmable data processing device generate means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0107] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means that implement the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0109] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0110] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0111] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0112] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0113] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0114] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0115] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple. For the relevant parts, reference can be made to the corresponding description in the method embodiment.

[0116] The above description is only for the embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various modifications and changes can be made to this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this specification.

Claims

1. A method for problem recommendation, characterized in that, including: Determine a number of historical questions raised by the user in the current session; According to each historical question, use a pre-trained question prediction model to determine the predicted question output by the question prediction model; wherein, the question prediction model is a large language model; Determine a first question that matches the predicted question from among a number of first questions included in a pre-constructed Q&A database, and use it as a candidate question; Determine a target question according to the candidate question, and recommend the target question to the user.

2. The method according to claim 1, characterized in that, According to each historical question, use a pre-trained question prediction model to determine a predicted question, specifically including: Determine a key question from each historical question; Input the key question into the pre-trained question prediction model to determine the predicted question output by the question prediction model.

3. The method according to claim 2, characterized in that Determine a key question from each historical question, specifically including: Screen each historical question to determine historical questions within a preset time range, and use them as second questions; Input the second question into the question prediction model to determine the key question output by the question prediction model.

4. The method according to claim 1, wherein Determine a first question that matches the predicted question from among a number of first questions included in a pre-constructed question database, and use it as a candidate question, specifically including: For each first question included in the pre-constructed Q&A database, determine the first similarity between the first question and the predicted question; According to the first similarity corresponding to each first question, sort the first questions in descending order of the first similarity to obtain a question sequence; According to the order of the question sequence, determine a specified number of first questions and use them as candidate questions that match the predicted question.

5. The method according to claim 1, characterized in that, Determine a target question according to the candidate question, specifically including: For each candidate question, determine the second similarity corresponding to the candidate question and each of the historical questions; Judge whether there is a second similarity that reaches a preset threshold among the second similarities corresponding to the candidate question; If so, do not use the candidate question as the target question; If not, use the candidate question as the target question.

6. The method according to claim 1, wherein Pre-train the question prediction model, specifically including: Obtain question data in a historical session; Use the questions within the first time period in the question data as the first samples, and use the questions within the second time period in the question data as the first annotations of the first samples; wherein, the first time period is earlier than the second time period; Input the first samples into the question prediction model to be trained to determine the prediction result output by the question prediction model to be trained; Train the question prediction model to be trained according to the prediction result and the first annotation.

7. The method according to claim 1, wherein The method further includes: In response to the user's click operation, determine the target question selected by the user and use it as the question to be replied; Determine the first reply corresponding to the question to be replied from among a number of first replies included in the Q&A database and use it as the target reply; Display the target reply to the user.

8. An apparatus for problem recommendation, characterized in that, including: A determination module for determining a number of historical questions raised by the user in the current session; A prediction module, configured to determine a predicted question output by the question prediction model according to each historical question by using a pre-trained question prediction model; wherein, the question prediction model is a large language model; A screening module, configured to determine a first question that matches the predicted question from several first questions included in a pre-constructed Q&A database and use it as a candidate question; A recommendation module, configured to determine a target question according to the candidate question and recommend the target question to the user.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 7 above is implemented.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, the method described in any one of claims 1 to 7 above is implemented.

Images