Recommendation problem processing based on large model and model training method and device

Abstract

The present disclosure provides a method and device for processing recommendation questions and training models based on a large model, and relates to artificial intelligence fields such as large models, intelligent recommendations, deep learning, and natural language processing. The method may include: generating an answer corresponding to a target question and M first candidate recommendation questions, where M is a positive integer greater than 1; generating estimated click results for each first candidate recommendation question using a recommendation question model according to the target question and the answer; sorting each first candidate recommendation question according to the estimated click results, and determining the second candidate recommendation question in the top N positions among the sorted first candidate recommendation questions, where N is a positive integer and less than M; and outputting the answer and the second candidate recommendation question.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to large-model-based recommendation problem processing and model training methods and devices in the fields of large models, intelligent recommendation, deep learning, and natural language processing. Background Art

[0002] With the development of technology, human-computer dialogue functions have been used more and more widely. For example, users can open an application with human-computer dialogue function, enter a text-based question (query), and obtain a response (answer) generated and displayed in text form, thereby helping users solve problems. Summary of the invention

[0003] The present invention provides a method and device for processing recommendation problems based on a large model and training the model.

[0004] A method for processing a recommendation problem based on a large model, comprising:

[0005] Generate the answer corresponding to the target question and M first candidate recommendation questions, where M is a positive integer greater than 1;

[0006] Generate estimated click results of each of the first candidate recommended questions respectively using a recommended question model according to the target question and the answer;

[0007] Sorting the first candidate recommended questions according to the estimated click results, and determining the second candidate recommended questions that are in the top N positions among the sorted first candidate recommended questions, where N is a positive integer and is less than M;

[0008] The answer and the second candidate recommendation question are output.

[0009] A training method for a recommendation question model, comprising:

[0010] Acquire historical point expansion data, the historical point expansion data including: a target question, an answer corresponding to the target question, and a second candidate recommended question corresponding to the target question;

[0011] Acquire training samples according to the historical point spread data;

[0012] A recommendation question model is trained according to the training samples, and the recommendation question model is used to generate estimated click results of M first candidate recommendation questions to be recommended, respectively, where M is a positive integer greater than 1.

[0013] A recommendation problem processing device based on a large model, comprising: a first generation module, a second generation module and a recommendation processing module;

[0014] The first generating module is used to generate an answer corresponding to the target question and M first candidate recommendation questions, where M is a positive integer greater than 1;

[0015] The second generating module is used to generate estimated click results of each of the first candidate recommended questions respectively according to the target question and the answer by using the recommended question model;

[0016] The recommendation processing module is used to sort the first candidate recommended questions according to the estimated click results, and determine the second candidate recommended questions in the top N positions among the sorted first candidate recommended questions, and output the answer and the second candidate recommended questions, where N is a positive integer and is less than M.

[0017] A training device for a recommendation question model, comprising: a data acquisition module and a model training module;

[0018] The data acquisition module is used to acquire historical point expansion data, wherein the historical point expansion data includes: a target question, an answer corresponding to the target question, and a second candidate recommended question corresponding to the target question;

[0019] The model training module is used to obtain training samples based on the historical point display data, and train a recommendation question model based on the training samples. The recommendation question model is used to generate estimated click results for M first candidate recommendation questions to be recommended, where M is a positive integer greater than 1.

[0020] An electronic device, comprising:

[0021] at least one processor; and

[0022] a memory communicatively connected to the at least one processor; wherein,

[0023] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.

[0024] A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the method as described above.

[0025] A computer program product comprises a computer program / instruction, wherein the computer program / instruction implements the method described above when executed by a processor.

[0026] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0028] Figure 1 A flowchart of an embodiment of a method for processing a recommendation problem based on a large model described in the present disclosure;

[0029] Figure 2 A schematic diagram of a display method of the target question, the answer corresponding to the target question, and the corresponding second candidate recommended question described in the present disclosure;

[0030] Figure 3 A flowchart of an embodiment of a training method for a recommendation question model disclosed in the present invention;

[0031] Figure 4 It is a schematic diagram of the composition structure of an embodiment 400 of a recommendation question processing device based on a large model according to the present disclosure;

[0032] Figure 5 It is a schematic diagram of the composition structure of an embodiment 500 of a training device for a recommendation question model according to the present disclosure;

[0033] Figure 6 A schematic block diagram of an electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0034] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0035] In addition, it should be understood that the term "and / or" in this article is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0036] Figure 1 Flow chart of an embodiment of the method for processing recommendation problems based on a large model described in the present disclosure. Figure 1 As shown, the following specific implementation methods are included.

[0037] In step 101, an answer corresponding to a target question and M first candidate recommended questions are generated, where M is a positive integer greater than 1.

[0038] In step 102, based on the target question and the answer, the recommended question model is used to generate the estimated click results of each first candidate recommended question.

[0039] In step 103, the first candidate recommended questions are sorted according to the estimated click results, and the second candidate recommended questions in the top N positions are determined among the sorted first candidate recommended questions, and the answers and the second candidate recommended questions are output, where N is a positive integer and is less than M.

[0040] For the sake of distinction, the question to be answered input by the user is called the target question. In the traditional method, the corresponding answer will be generated and displayed for the target question input by the user. In addition, multiple recommended questions are usually generated and some are randomly selected for display to guide the user to click on one of the recommended questions to increase the number of conversation rounds, etc. However, since the recommended questions displayed are randomly selected, the actual effect is usually not ideal.

[0041] By adopting the solution described in the above method embodiment, the displayed second candidate recommendation question is selected from each first candidate recommendation question based on the estimated click result, so that the displayed second candidate recommendation question can better match the user's needs, thereby improving the accuracy of the recommendation results, etc.

[0042] The target question may be a question input by the user in text form, or may be a question obtained by converting the voice input by the user into text form. For the target question, corresponding answers and M first candidate recommended questions may be generated respectively, and the specific value of M may be determined according to actual needs. There is no limit to the way of generating answers and first candidate recommended questions.

[0043] For example, the target question is "winter outfit in 20 words", and the corresponding response to the target question may be "warm and fashionable outfit in cold winter, down sweater with long boots, simple yet elegant", and the first candidate recommendation questions may include: "do you have any other suggestions for winter outfits", "describe winter outfits more vividly", and "are there any other 20-word descriptions of winter outfits recommended", etc.

[0044] For the M first candidate recommended questions, estimated click results of each first candidate recommended question may be generated according to the target question and the answer corresponding to the target question.

[0045] In some embodiments of the present disclosure, for any first candidate recommendation question, the following processing can be performed respectively: the target question, the answer corresponding to the target question and the first candidate recommendation question are input into the recommendation question model to obtain the estimated click result of the first candidate recommendation question as output.

[0046] For example, taking the first candidate recommendation question "Describe winter outfits more vividly" as an example, the target question "Winter outfits in 20 words", the corresponding answer to the target question "Warm and fashionable outfits in cold winter, down sweaters with long boots, simple yet elegant", and the first candidate recommendation question "Describe winter outfits more vividly" can be spliced ​​in a predetermined way as the input of the recommendation question model, thereby obtaining the estimated click results output by the recommendation question model.

[0047] The recommendation question model can be a large model, which refers to a large language model (LLM, Large Language Model), which is a deep learning model trained with a large amount of text data, and can generate natural language text or understand the meaning of language text, etc. In the solution disclosed in the present invention, the estimated click result corresponding to the first candidate recommendation question is generated by combining the target question, the answer corresponding to the target question, and the first candidate recommendation question itself, thereby improving the accuracy of the generated estimated click result, etc.

[0048] After respectively obtaining the estimated click results of the M first candidate recommended questions, the M first candidate recommended questions may be sorted according to the estimated click results.

[0049] In some embodiments of the present disclosure, the estimated click result may include: an estimated click parameter, the larger the value of the estimated click parameter, the greater the probability that the corresponding first candidate recommended question will be clicked; accordingly, the method of sorting each first candidate recommended question according to the estimated click result may include: sorting each first candidate recommended question in order from large to small according to the value of the estimated click parameter.

[0050] The estimated click parameter may be an estimated click score, and the value of the estimated click score may be between 0 and 1.

[0051] Alternatively, in some embodiments of the present disclosure, the method of sorting each first candidate recommended question according to the estimated click results may include: performing the following processing for each first candidate recommended question: obtaining the correlation parameter between the first candidate recommended question and the target question, and determining the comprehensive parameter of the first candidate recommended question according to the correlation parameter and the estimated click parameter; and sorting each first candidate recommended question in order of the value of the comprehensive parameter from large to small.

[0052] The relevance parameter may refer to a relevance score. Taking the first candidate recommendation question a as an example, there is no restriction on how to obtain the relevance score between the first candidate recommendation question a and the target question. For example, the semantic similarity between the first candidate recommendation question a and the target question may be obtained, and the semantic similarity may be used as the relevance score. The value of the relevance score may also be between 0 and 1.

[0053] In addition, the comprehensive parameter can be a comprehensive score. After obtaining the relevance score and the estimated click score of the first candidate recommendation question a, the relevance score and the estimated click score can be multiplied by the corresponding weights, respectively, and the two products can be added. The sum is used as the comprehensive score of the first candidate recommendation question a. The specific value of each weight can be determined according to actual needs.

[0054] In practical applications, any of the above methods can be used to sort the first candidate recommended questions, which is very flexible and convenient. Preferably, the latter method can be used. Since the obtained comprehensive parameters simultaneously integrate the estimated click information of the first candidate recommended questions and the correlation information with the target question, accordingly, sorting the first candidate recommended questions based on the comprehensive parameters can improve the accuracy of the sorting results.

[0055] After ranking the first candidate recommended questions, the first candidate recommended questions in the top N positions after ranking can be determined as the second candidate recommended questions, and the specific value of N can be determined according to actual needs, such as 3. Further, the answer corresponding to the target question and the second candidate recommended questions can be displayed to the user.

[0056] Figure 2 Schematic diagram of the display method of the target question, the answer corresponding to the target question and the corresponding second candidate recommended question described in the present disclosure. Figure 2 As shown, if necessary, the user can click on a displayed second candidate recommended question. Accordingly, the clicked second candidate recommended question can be used as the target question for the next round of dialogue, and the relevant processing described in the present disclosure can be repeated.

[0057] In addition, it can be seen from the above introduction that the implementation of the solution described in the present disclosure needs to rely on a pre-trained recommendation question model. The specific training method of the recommendation question model is described below.

[0058] Figure 3 Flow chart of an embodiment of a training method for a recommendation question model described in the present disclosure. Figure 3 As shown, the following specific implementation methods are included.

[0059] In step 301, historical point expansion data is obtained, and the historical point expansion data includes: a target question, an answer corresponding to the target question, and a second candidate recommended question corresponding to the target question.

[0060] In step 302, training samples are obtained based on historical point display data, and a recommendation question model is trained based on the training samples. The recommendation question model is used to generate estimated click results for M first candidate recommendation questions to be recommended, where M is a positive integer greater than 1.

[0061] By adopting the scheme described in the above method embodiment, training samples can be obtained according to historical point display data, and the recommendation question model can be trained according to the training samples. Accordingly, after respectively generating the answer corresponding to the target question and M first candidate recommended questions, the trained recommendation question model can be used to generate the estimated click results of each first candidate recommended question, and the second candidate recommended question can be determined from each first candidate recommended question based on the estimated click results, and then the second candidate recommended question can be displayed to the user together with the generated answer, so that the displayed second candidate recommended question can better match the user's needs, thereby improving the accuracy of the recommendation results, etc.

[0062] The historical point expansion data may be real data obtained online, which may include the target question input by the user, the answer corresponding to the generated target question, and the second candidate recommended question displayed together with the answer. It may also include whether the user clicks on the displayed second candidate recommended question and which second candidate recommended question the user clicks on.

[0063] In some embodiments of the present disclosure, before obtaining training samples based on historical point spread data, the following optimization processing can also be performed: in response to determining that the number of characters included in the response is greater than L, where L is a positive integer greater than 1, the characters in the response starting from the L+1th character are deleted.

[0064] The specific value of L can be determined according to actual needs, such as 100. That is to say, if the number of characters included in the response of a certain historical point extension data is greater than 100, only the first 100 characters can be retained and the remaining characters can be deleted, thereby preventing the response from being too long and affecting subsequent processing efficiency and processing accuracy.

[0065] After completing the above optimization process, the historical point spread data can be used to obtain training samples. In some embodiments of the present disclosure, positive samples and negative samples can be obtained respectively according to the historical point spread data, wherein the negative samples may include: two different types of negative samples, namely, first-class negative samples and second-class negative samples. Accordingly, the recommendation problem model can be subsequently trained according to the positive samples, the first-class negative samples, and the second-class negative samples.

[0066] Through the above processing, the constructed sample types can be enriched, thereby improving the learning ability of the model, that is, improving the training effect of the model, etc.

[0067] In some embodiments of the present disclosure, positive samples come from the same historical point display data, and the positive samples include: a target question, an answer corresponding to the target question, and a second candidate recommended question that is clicked among the second candidate recommended questions corresponding to the target question.

[0068] For example, assuming that a certain historical point expansion data b includes: target question b, answer b corresponding to target question b, and three second candidate recommended questions displayed together with answer b, namely second candidate recommended question 1, second candidate recommended question 2 and second candidate recommended question 3, among which the user clicks on second candidate recommended question 2. Then a positive sample can be constructed based on the historical point expansion data b, such as the positive sample may include: target question b, answer b and second candidate recommended question 2.

[0069] In addition, in some embodiments of the present disclosure, the first category of negative samples comes from two different historical point expansion data, and the first category of negative samples may include: a target question and an answer corresponding to the target question from the same first historical point expansion data, and a second candidate recommended question that is not clicked and comes from other second historical point expansion data other than the first historical point expansion data; the second category of negative samples comes from the same historical point expansion data, and the second category of negative samples may include: a target question, an answer corresponding to the target question, and a second candidate recommended question that is not clicked among the second candidate recommended questions corresponding to the target question.

[0070] For example, assuming that another historical point expansion data c includes: target question c, answer c corresponding to target question c, and three second candidate recommended questions displayed together with answer c, namely second candidate recommended question 11, second candidate recommended question 22 and second candidate recommended question 33, among which the user clicks on second candidate recommended question 11, then a first-class negative sample can be constructed in combination with the historical point expansion data b and the historical point expansion data c, such as the first-class negative sample may include: target question b, answer b and second candidate recommended question 33.

[0071] For another example, a second type of negative sample can be constructed based on the historical point expansion data b, and the second type of negative sample may include: target question b, answer b and second candidate recommended question 1. Similarly, a second type of negative sample can be constructed based on the historical point expansion data c, and the second type of negative sample may include: target question c, answer c and second candidate recommended question 22.

[0072] It can be seen that through the above processing, the required various training samples can be constructed efficiently and accurately, thus laying a good foundation for subsequent model training.

[0073] In addition, whether it is a positive sample or a negative sample, it includes a target question, the answer corresponding to the target question, and a second candidate recommendation question. Accordingly, each training sample can be expressed as follows: <que>Query content <ans>Answer content <sug>sug content, where " <que>query content" indicates the target question, <ans>The answer content" indicates the answer to the target question, <sug>sug content" indicates the second candidate recommendation question.

[0074] In addition, in some embodiments of the present disclosure, the ratio of the number of positive samples (pos) to the number of negative samples (neg) may be equal to a preset first value, and the ratio of the number of first-class negative samples (easy) to the number of second-class negative samples (hard) may be equal to a preset second value.

[0075] The specific values ​​of the first value and the second value can be determined according to actual needs. For example, pos:neg=1:9, easy:hard=1:5.

[0076] By reasonably setting the ratio of the number of positive samples to negative samples and the ratio of the number of first-category negative samples to second-category negative samples, the training effect of the model can be effectively improved.

[0077] The specific structure of the recommendation question model in the scheme disclosed in the present invention is not limited, as long as the corresponding function can be realized. As a possible implementation method, for any training sample, the target question, the answer and the second candidate recommendation question can be spliced ​​in a predetermined manner as the input text of the recommendation question model. Accordingly, the processing performed by the recommendation question model may include: using a tokenizer to perform word segmentation on the input text and obtain the initial vector of each word segmentation result, and then using the encoder of the multilingual version of the T5 (MT5, Multilingual T5) model to convert and encode the initial vector to obtain an update vector, T5 refers to the text-to-text transfer transformer, and further, the update vector can be reduced in dimension and input into a classification head composed of two fully connected layers and corresponding activation functions, and finally output an estimated click score between 0 and 1. In addition, when training, the estimated click score of the positive sample can be considered as 1, and the estimated click score of the negative sample can be considered as 0.

[0078] After completing the training of the recommendation question model using the constructed training samples, the recommendation question model can be applied to actual question recommendation scenarios, such as Figure 1 In the illustrated embodiment, it is used to generate an estimated click result for each first candidate recommendation question, and the estimated click result may be an estimated click score.

[0079] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present disclosure is not limited by the described order of actions, because according to the present disclosure, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present disclosure. In addition, for parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0080] The above is an introduction to the method embodiment. The following is a further explanation of the scheme disclosed in the present invention through an apparatus embodiment.

[0081] Figure 4 FIG. 4 is a schematic diagram of the composition structure of an embodiment 400 of a recommendation problem processing device based on a large model according to the present disclosure. Figure 4 As shown, it includes: a first generating module 401, a second generating module 402 and a recommendation processing module 403.

[0082] The first generating module 401 is used to generate an answer corresponding to the target question and M first candidate recommended questions, where M is a positive integer greater than 1.

[0083] The second generating module 402 is used to generate the estimated click results of each first candidate recommended question respectively according to the target question and the answer by using the recommended question model.

[0084] The recommendation processing module 403 is used to sort the first candidate recommended questions according to the estimated click results, and determine the second candidate recommended questions in the top N positions among the sorted first candidate recommended questions, and output the answer and the second candidate recommended question, where N is a positive integer and is less than M.

[0085] By adopting the scheme described in the above-mentioned device embodiment, the displayed second candidate recommendation question is selected from each first candidate recommendation question based on the estimated click result, so that the displayed second candidate recommendation question can better match the user's needs, thereby improving the accuracy of the recommendation results, etc.

[0086] For the target question, the first generation module 401 can generate corresponding answers and M first candidate recommended questions respectively. Then, for the M first candidate recommended questions, the second generation module 402 can generate estimated click results for each first candidate recommended question according to the target question and the answers corresponding to the target question.

[0087] In some embodiments of the present disclosure, for any first candidate recommendation question, the second generation module 402 can perform the following processing respectively: input the target question, the answer corresponding to the target question and the first candidate recommendation question into the recommendation question model, and obtain the estimated click result of the first candidate recommendation question as output.

[0088] After respectively obtaining the estimated click results of the M first candidate recommended questions, the recommendation processing module 403 may sort the M first candidate recommended questions according to the estimated click results.

[0089] In some embodiments of the present disclosure, the estimated click result may include: an estimated click parameter, the larger the value of the estimated click parameter, the greater the probability that the corresponding first candidate recommended question will be clicked; accordingly, the recommendation processing module 403 may sort the first candidate recommended questions according to the estimated click result, which may include: sorting the first candidate recommended questions in order from large to small according to the value of the estimated click parameter.

[0090] Alternatively, in some embodiments of the present disclosure, the method in which the recommendation processing module 403 sorts each first candidate recommended question according to the estimated click results may include: performing the following processing for each first candidate recommended question: obtaining the correlation parameter between the first candidate recommended question and the target question, and determining the comprehensive parameter of the first candidate recommended question according to the correlation parameter and the estimated click parameter; and sorting each first candidate recommended question in order of the value of the comprehensive parameter from large to small.

[0091] Figure 5 FIG. 5 is a schematic diagram of the structure of an embodiment of a training device 500 for a recommendation question model according to the present disclosure. Figure 5 As shown, it includes: a data acquisition module 501 and a model training module 502.

[0092] The data acquisition module 501 is used to acquire historical point expansion data, where the historical point expansion data includes: a target question, an answer corresponding to the target question, and a second candidate recommended question corresponding to the target question.

[0093] The model training module 502 is used to obtain training samples based on historical point display data, and train a recommendation question model based on the training samples. The recommendation question model is used to generate estimated click results for M first candidate recommendation questions to be recommended, where M is a positive integer greater than 1.

[0094] By adopting the scheme described in the above-mentioned device embodiment, training samples can be obtained according to historical point display data, and the recommendation question model can be trained according to the training samples. Accordingly, after respectively generating the answer corresponding to the target question and M first candidate recommended questions, the trained recommendation question model can be used to generate the estimated click results of each first candidate recommended question, and the second candidate recommended question can be determined from each first candidate recommended question based on the estimated click results, and then the second candidate recommended question can be displayed to the user together with the generated answer, so that the displayed second candidate recommended question can better match the user's needs, thereby improving the accuracy of the recommendation results, etc.

[0095] In some embodiments of the present disclosure, before acquiring training samples based on historical point spread data, the data acquisition module 501 may also perform the following optimization processing: in response to determining that the number of characters included in the response of the historical point spread data is greater than L, where L is a positive integer greater than 1, delete the characters in the response starting from the L+1th character.

[0096] After completing the above optimization process, the model training module 502 can use the historical point spread data to obtain training samples. In some embodiments of the present disclosure, the model training module 502 can obtain positive samples and negative samples based on the historical point spread data, wherein the negative samples may include two different types of negative samples: first-class negative samples and second-class negative samples. Accordingly, the recommendation problem model can be trained based on the positive samples, the first-class negative samples, and the second-class negative samples.

[0097] In some embodiments of the present disclosure, the ratio of the number of positive samples to the number of negative samples may be equal to a preset first value, and the ratio of the number of first-category negative samples to the number of second-category negative samples may be equal to a preset second value.

[0098] In addition, in some embodiments of the present disclosure, the positive samples come from the same historical point display data, and the positive samples may include: the target question, the answer corresponding to the target question, and the second candidate recommended question that is clicked in the second candidate recommended question corresponding to the target question.

[0099] In some embodiments of the present disclosure, the first category of negative samples comes from two different historical point expansion data, and the first category of negative samples may include: a target question from the same first historical point expansion data, an answer corresponding to the target question, and a second candidate recommended question from other second historical point expansion data other than the first historical point expansion data and not clicked; the second category of negative samples comes from the same historical point expansion data, and the second category of negative samples may include: the target question, an answer corresponding to the target question, and a second candidate recommended question that was not clicked among the second candidate recommended questions corresponding to the target question.

[0100] Figure 4 and Figure 5 The specific working process of the illustrated device embodiment can refer to the relevant description in the aforementioned method embodiment and will not be described in detail.

[0101] The solution disclosed in this disclosure can be applied to the field of artificial intelligence, especially to the fields of large models, intelligent recommendations, deep learning, and natural language processing. Artificial intelligence is a discipline that studies how computers can simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.). It has both hardware-level and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing. Artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, as well as machine learning / deep learning, big data processing technology, knowledge graph technology, and other major directions.

[0102] The questions, answers, and recommended questions in the embodiments described in the present disclosure are not targeted at a specific user and do not reflect the personal information of a specific user. In addition, the executor of the method described in the present disclosure may obtain the questions in various public, legal, and compliant ways, such as obtaining them from the user with the user's authorization.

[0103] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0104] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0105] Figure 6 A schematic block diagram of an electronic device 600 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0106] like Figure 6 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 to a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the electronic device 600 can also be stored. The computing unit 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0107] Multiple components in the electronic device 600 are connected to the I / O interface 605, including: an input unit 606, such as a keyboard, a mouse, etc.; an output unit 607, such as various types of displays, speakers, etc.; a storage unit 608, such as a disk, an optical disk, etc.; and a communication unit 609, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 609 allows the electronic device 600 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0108] The computing unit 601 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI, Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSP, Digital Signal Processing), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 601 performs the various methods and processes described above, such as the methods described in the present disclosure. For example, in some embodiments, the methods described in the present disclosure may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 608. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 600 via the ROM 602 and / or the communication unit 609. When the computer program is loaded into the RAM 603 and executed by the computing unit 601, one or more steps of the methods described in the present disclosure may be executed. Alternatively, in other embodiments, the computing unit 601 may be configured to execute the method described in the present disclosure in any other appropriate manner (for example, by means of firmware).

[0109] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard parts (ASSPs), system on chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs, which can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor, which can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0110] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0111] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (EPROM, Electronically Programmable Read-Only Memory), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM, Compact Disc Read-Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0112] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or a liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0113] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0114] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0115] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0116] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.< / sug> < / ans> < / que> < / sug> < / ans> < / que>

Claims

1. A method for processing recommendation problems based on a large model, comprising: Generate the answer corresponding to the target question and M first candidate recommendation questions, where M is a positive integer greater than 1; Generate estimated click results of each of the first candidate recommended questions respectively using a recommended question model according to the target question and the answer; Sorting the first candidate recommended questions according to the estimated click results, and determining the second candidate recommended questions that are in the top N positions among the sorted first candidate recommended questions, where N is a positive integer and is less than M; The answer and the second candidate recommendation question are output.

2. The method according to claim 1, wherein: The generating, according to the target question and the answer, respectively estimated click results of each of the first candidate recommended questions by using the recommended question model comprises: The target question, the answer and each of the first candidate recommendation questions are input into the recommendation question model to obtain the output estimated click result.

3. The method according to claim 1 or 2, wherein: The estimated click result includes: an estimated click parameter, wherein a larger value of the estimated click parameter indicates a greater probability that the corresponding first candidate recommended question will be clicked; The sorting of the first candidate recommendation questions according to the estimated click results includes: The first candidate recommendation questions are sorted in descending order according to the values ​​of the estimated click parameters.

4. The method according to claim 1 or 2, wherein: The estimated click result includes: an estimated click parameter, wherein the larger the value of the estimated click parameter is, the greater the probability that the corresponding first candidate recommended question is clicked is; The sorting of the first candidate recommendation questions according to the estimated click results includes: Respectively obtaining correlation parameters between each of the first candidate recommendation questions and the target question; Determine a comprehensive parameter of each of the first candidate recommendation questions according to the relevance parameter and the estimated click parameter; The first candidate recommendation questions are sorted in descending order according to the values ​​of the comprehensive parameter.

5. A training method for a recommendation question model, comprising: Acquire historical point expansion data, the historical point expansion data including: a target question, an answer corresponding to the target question, and a second candidate recommended question corresponding to the target question; Acquire training samples according to the historical point spread data; A recommendation question model is trained according to the training samples, and the recommendation question model is used to generate estimated click results of M first candidate recommendation questions to be recommended, respectively, where M is a positive integer greater than 1.

6. The method according to claim 5, before acquiring training samples according to the historical point spread data, the method further comprises: In response to determining that the number of characters included in the response is greater than L, characters starting from the L+1th character in the response are deleted, where L is a positive integer greater than 1.

7. The method according to claim 5 or 6, wherein: The acquiring of training samples according to the historical point spread data comprises: acquiring positive samples and negative samples according to the historical point spread data; wherein the negative samples comprise: two different types of negative samples: first type of negative samples and second type of negative samples; The training of the recommendation problem model according to the training samples includes: training the recommendation problem model according to the positive samples, the first category of negative samples and the second category of negative samples.

8. The method according to claim 7, wherein: The ratio of the number of the positive samples to the number of the negative samples is equal to a preset first value, and the ratio of the number of the first category of negative samples to the number of the second category of negative samples is equal to a preset second value.

9. The method according to claim 7, wherein: The positive sample comes from the same piece of historical point display data, and the positive sample includes: the target question, the answer corresponding to the target question, and the second candidate recommended question that is clicked among the second candidate recommended questions corresponding to the target question.

10. The method according to claim 7, wherein: The first type of negative samples comes from two different historical point display data, and the first type of negative samples includes: the target question from the same first historical point display data, the answer corresponding to the target question, and the second candidate recommendation question from other second historical point display data other than the first historical point display data and not clicked; The second type of negative samples come from the same historical point display data, and the second type of negative samples include: the target question, the answer corresponding to the target question, and the second candidate recommended questions that are not clicked in the second candidate recommended questions corresponding to the target question.

11. A recommendation problem processing device based on a large model, comprising: A first generating module, a second generating module and a recommendation processing module; The first generating module is used to generate an answer corresponding to the target question and M first candidate recommendation questions, where M is a positive integer greater than 1; The second generating module is used to generate estimated click results of each of the first candidate recommended questions respectively according to the target question and the answer by using the recommended question model; The recommendation processing module is used to sort the first candidate recommended questions according to the estimated click results, and determine the second candidate recommended questions in the top N positions among the sorted first candidate recommended questions, and output the answer and the second candidate recommended questions, where N is a positive integer and is less than M.

12. The device according to claim 11, wherein The second generation module inputs the target question, the answer and each of the first candidate recommendation questions into the recommendation question model to obtain the output estimated click result.

13. The device according to claim 11 or 12, wherein: The estimated click result includes: an estimated click parameter, wherein a larger value of the estimated click parameter indicates a greater probability that the corresponding first candidate recommended question will be clicked; The recommendation processing module sorts the first candidate recommendation questions in descending order according to the values ​​of the estimated click parameters.

14. The device according to claim 11 or 12, wherein: The estimated click result includes: an estimated click parameter, wherein a larger value of the estimated click parameter indicates a greater probability that the corresponding first candidate recommended question will be clicked; The recommendation processing module respectively obtains the correlation parameters between each of the first candidate recommendation questions and the target question, determines the comprehensive parameters of each of the first candidate recommendation questions according to the correlation parameters and the estimated click parameters, and sorts each of the first candidate recommendation questions in descending order according to the values ​​of the comprehensive parameters.

15. A training device for a recommendation question model, comprising: Data acquisition module and model training module; The data acquisition module is used to acquire historical point expansion data, wherein the historical point expansion data includes: a target question, an answer corresponding to the target question, and a second candidate recommended question corresponding to the target question; The model training module is used to obtain training samples based on the historical point display data, and train a recommendation question model based on the training samples. The recommendation question model is used to generate estimated click results for M first candidate recommendation questions to be recommended, where M is a positive integer greater than 1.

16. The device according to claim 15, wherein: The data acquisition module is further used to, in response to determining that the number of characters included in the response is greater than L, delete characters starting from the L+1th character in the response, where L is a positive integer greater than 1.

17. The device according to claim 15 or 16, wherein: The model training module obtains positive samples and negative samples according to the historical point display data, wherein the negative samples include two different types of negative samples: first type of negative samples and second type of negative samples, and trains the recommendation problem model according to the positive samples, the first type of negative samples and the second type of negative samples.

18. The device according to claim 17, wherein: The ratio of the number of the positive samples to the number of the negative samples is equal to a preset first value, and the ratio of the number of the first category of negative samples to the number of the second category of negative samples is equal to a preset second value.

19. The device according to claim 17, wherein: The positive sample comes from the same piece of historical point display data, and the positive sample includes: the target question, the answer corresponding to the target question, and the second candidate recommended question that is clicked among the second candidate recommended questions corresponding to the target question.

20. The device according to claim 17, wherein: The first type of negative samples comes from two different historical point display data, and the first type of negative samples includes: the target question from the same first historical point display data, the answer corresponding to the target question, and the second candidate recommendation question from other second historical point display data other than the first historical point display data and not clicked; The second type of negative samples come from the same historical point display data, and the second type of negative samples include: the target question, the answer corresponding to the target question, and the second candidate recommended questions that are not clicked in the second candidate recommended questions corresponding to the target question.

21. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 10.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to make a computer execute the method according to any one of claims 1 to 10.

23. A computer program product, comprising a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

Images