A multi-task recommendation method and device fusing prior information

Abstract

This application provides a multi-task recommendation method and apparatus that integrates prior information, relating to the field of artificial intelligence and also applicable to the financial field. The method includes: generating expert representation vectors for each expert and a user representation vector for the current questioning user based on embedded data corresponding to question-and-answer prior information; wherein the embedded data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user; determining multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector; and recommending the optimal answer corresponding to the current question information from all answers provided by the multiple experts. This application can integrate pre-stored question-and-answer prior information for multi-task recommendation, utilizing the strong correlation between data to help users quickly retrieve the optimal answer to their questions.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence and can be used in the financial field. Specifically, it is a multi-task recommendation method and device that integrates prior information. Background Technology

[0002] Information retrieval is a way for users to obtain information. In a narrow sense, information retrieval mainly refers to information search, which is the process by which users input their questions based on their needs and use search tools to find the information they need from a collection of information.

[0003] Based on the above, the financial sector has introduced an online Q&A platform. Information retrieval is one of its main functions. Ensuring that users can quickly and efficiently obtain the desired answers during information retrieval is crucial for enhancing user engagement and building a positive product reputation. In practical applications, the question-answering rate and resolution timeliness are important user experience indicators. A key technical challenge is finding the most likely experts and answers to solve a user's question within the existing expert knowledge base. Summary of the Invention

[0004] To address the problems in the existing technology, this application provides a multi-task recommendation method and apparatus that integrates prior information. It can integrate pre-stored question-and-answer prior information for multi-task recommendation and utilize the strong correlation between data to help users quickly retrieve the optimal answer to their questions.

[0005] To solve the above-mentioned technical problems, this application provides the following technical solution:

[0006] Firstly, this application provides a multi-task recommendation method that integrates prior information, including:

[0007] Based on the prior information of the questions and answers, an expert representation vector and a user representation vector corresponding to the current questioning user are generated. The prior information of the questions and answers includes the current question information of the current questioning user.

[0008] Based on the user representation vector and the expert representation vector, determine multiple experts corresponding to the current question information;

[0009] The optimal answer corresponding to the current question information is recommended from all the answers provided by the multiple experts.

[0010] Furthermore, the step of generating expert representation vectors for each expert and user representation vectors for the current question-asking user based on the embedded data corresponding to the question-answering prior information includes:

[0011] The full set of user question data and full set of expert answer data are input into a long short-term memory artificial neural network to obtain full set of question word vectors and full set of answer word vectors;

[0012] The user representation vector is constructed based on the question word vector corresponding to the current questioning user in the full set of question word vectors;

[0013] The expert representation vector is constructed based on the answer word vectors corresponding to each expert in the full set of answer word vectors.

[0014] Further, determining the multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector includes:

[0015] Calculate the similarity between the expert representation vector and the user representation vector for each expert to obtain the first similarity value;

[0016] The first similarity value is normalized and sorted to obtain multiple experts corresponding to the current question information.

[0017] Furthermore, recommending the optimal answer corresponding to the current question information from all the answers provided by the multiple experts includes:

[0018] Select the answer word vectors corresponding to all the answers provided by the multiple experts from the full set of answer word vectors;

[0019] Calculate the similarity between each answer word vector and the question word vector corresponding to the current questioning user to obtain a second similarity value;

[0020] The answer corresponding to the largest of the second similarity values ​​is selected as the optimal answer.

[0021] Furthermore, after recommending the optimal answer corresponding to the current question information from all the answers provided by the multiple experts, the method further includes:

[0022] A recommendation error loss function is constructed based on the first similarity value and the second similarity value;

[0023] The parameters of the long short-term memory artificial neural network are adjusted according to the recommended error loss function.

[0024] Further, the step of constructing a recommendation error loss function based on the normalized first similarity value and the second similarity value includes:

[0025] Calculate the expert recommendation error loss function based on the first similarity value;

[0026] The answer recommendation error loss function is calculated based on the second similarity value;

[0027] Construct a recommendation error loss function based on the expert recommendation error loss function and the answer recommendation error loss function.

[0028] Secondly, this application provides a multi-task recommendation device that integrates prior information, including:

[0029] The representation vector generation unit is used to generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the tracking data corresponding to the question-and-answer prior information; wherein, the tracking data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user;

[0030] The expert recommendation unit is used to determine multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector;

[0031] The answer recommendation unit is used to recommend the optimal answer corresponding to the current question information from all the answers provided by the multiple experts.

[0032] Further, the representation vector generation unit includes:

[0033] The full-volume word vector generation module is used to input the full-volume user question data and the full-volume expert answer data into the long short-term memory artificial neural network to obtain full-volume question word vectors and full-volume answer word vectors;

[0034] The user representation vector generation module is used to construct the user representation vector based on the question word vector corresponding to the current questioning user in the full set of question word vectors;

[0035] The expert representation vector generation module is used to construct the expert representation vector based on the answer word vectors corresponding to each expert in the full set of answer word vectors.

[0036] Furthermore, the expert recommendation unit includes:

[0037] The first similarity value calculation module is used to calculate the similarity between the expert representation vector and the user representation vector corresponding to each expert, and obtain the first similarity value.

[0038] The expert recommendation module is used to normalize and sort the first similarity value to obtain multiple experts corresponding to the current question information.

[0039] Furthermore, the answer recommendation unit includes:

[0040] The answer word vector selection module is used to select the answer word vectors corresponding to all the answers provided by the multiple experts from the full set of answer word vectors;

[0041] The second similarity value calculation module is used to calculate the similarity between each answer word vector and the question word vector corresponding to the current questioning user, and obtain the second similarity value;

[0042] The answer recommendation module is used to select the answer corresponding to the largest similarity value among the second similarity values ​​as the optimal answer.

[0043] Furthermore, the multi-task recommendation device that integrates prior information also includes:

[0044] The recommendation loss function construction unit is used to construct a recommendation error loss function based on the first similarity value and the second similarity value.

[0045] The network parameter adjustment unit is used to adjust the parameters of the long short-term memory artificial neural network according to the recommendation error loss function.

[0046] Furthermore, the recommendation loss function construction unit includes:

[0047] An expert loss function construction module is used to calculate the expert recommendation error loss function based on the first similarity value;

[0048] The answer loss function construction module is used to calculate the answer recommendation error loss function based on the second similarity value;

[0049] The recommendation loss function construction module is used to construct a recommendation error loss function based on the expert recommendation error loss function and the answer recommendation error loss function.

[0050] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the multi-task recommendation method that integrates prior information.

[0051] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-task recommendation method that integrates prior information.

[0052] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the multi-task recommendation method that integrates prior information.

[0053] To address the problems in existing technologies, the multi-task recommendation method and apparatus that integrates prior information provided in this application can capture the strong correlation between questions and answers through a prototype network, tailored to the application scenario and business characteristics. It can jointly train the information retrieval and recommendation expert tasks, narrow down the candidate answer set during the recommendation expert process, and ultimately quickly retrieve the optimal answer to the user's question, thereby improving the user's question resolution rate and achieving good results in practical applications. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of the multi-task recommendation method that integrates prior information in the embodiments of this application;

[0056] Figure 2 This is a flowchart illustrating the generation of representation vectors in an embodiment of this application;

[0057] Figure 3 This is a flowchart illustrating the process of determining experts in the embodiments of this application;

[0058] Figure 4 This is a flowchart illustrating the determination of the optimal answer in the embodiments of this application;

[0059] Figure 5 This is one of the flowcharts for adjusting neural network parameters in the embodiments of this application;

[0060] Figure 6 This is the second flowchart of adjusting neural network parameters in the embodiments of this application;

[0061] Figure 7 This is one of the structural diagrams of a multi-task recommendation device that integrates prior information in the embodiments of this application;

[0062] Figure 8 This is a structural diagram of the representation vector generation unit in the embodiments of this application;

[0063] Figure 9 This is a structural diagram of the expert recommendation unit in the embodiments of this application;

[0064] Figure 10 This is a structural diagram of the answer recommendation unit in an embodiment of this application;

[0065] Figure 11 This is the second structural diagram of the multi-task recommendation device that integrates prior information in the embodiments of this application;

[0066] Figure 12 This is a structural diagram of the recommended loss function construction unit in the embodiments of this application;

[0067] Figure 13 This is a schematic diagram of the structure of the electronic device in the embodiments of this application. Detailed Implementation

[0068] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0069] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0070] In one embodiment, see Figure 1 To enable multi-task recommendation by integrating pre-stored question-and-answer prior information and leveraging the strong correlations between data to help users quickly retrieve the optimal answer to their questions, this application provides a multi-task recommendation method that integrates prior information, including:

[0071] S101: Generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the tracking data corresponding to the question-and-answer prior information; wherein, the tracking data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user;

[0072] S102: Determine multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector;

[0073] S103: Recommend the optimal answer corresponding to the current question information from all the answers provided by the multiple experts.

[0074] It is understood that the embodiments of this application can be implemented based on online Q&A platforms introduced in the financial field. An online Q&A platform can be understood as a platform or system that allows users to ask questions online and experts to answer them online. In such platforms, both users and experts can register and log in beforehand to ask and answer questions.

[0075] Unlike general information retrieval platforms, online question-and-answer platforms can be used for mutual discussion and assistance in financial business. Users' questions need to be resolved in the shortest possible time. Existing methods for retrieving answers generally separate the tasks of information retrieval and expert recommendation, which is clearly detrimental to problem-solving efficiency. To address this issue, this application proposes a multi-task recommendation method for online question-and-answer platforms. This method leverages the strong correlation between question-and-answer pairs, using the prototype representation of a question (or answer) raised by the same user (or expert) as the representation of that user (or expert), simultaneously completing both information retrieval and expert recommendation tasks.

[0076] In the online Q&A platform, there are two roles: user c and expert e, forming user set C and expert set E respectively. Each user c corresponds to a set of questions Q. c Each expert e corresponds to a set of answers A. e All Q c The set of problems Q consists of all A's. e This constitutes the overall set of answers A. Given a problem q∈Q, there exist one or more answers (also called solutions) a∈A that can solve the problem.

[0077] In the application scenario of this application embodiment, the same user's question q∈Q c The answer a∈A belongs to the same expert e This information is highly relevant and can be integrated into the text representation process to improve the accuracy of information retrieval. The recommended expert is e. a When user c raises a new question q', experts who can solve the problem will be recommended.

[0078] In practical applications, after a period of operation on a crowdsourced Q&A online communication platform, historical data, known as prior information, is accumulated. Using event tracking technology (event data), this prior information can be obtained, and then an artificial neural network can be used to generate expert representation vectors for each expert and user representation vectors for the current questioner. The expert representation vector represents all questions the expert has answered historically; the user representation vector represents all questions the user has asked historically. Next, by analyzing the correlation between the user representation vector and the expert representation vector, multiple experts capable of answering the current question can be identified. Finally, the optimal answer for the current question is recommended from all the answers provided by these experts.

[0079] As can be seen from the above description, the multi-task recommendation method that integrates prior information provided in this application can capture the strong correlation between questions and answers through a prototype network, and jointly train the information retrieval and recommendation expert tasks according to the application scenario and business characteristics. In the process of recommending experts, the candidate answer set is narrowed down, and finally the optimal answer to the question asked by the user is retrieved quickly, which improves the problem-solving rate of users and has achieved good results in practical applications.

[0080] The following provides a detailed explanation of steps S101 to S103.

[0081] Step S101: Generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the tracking data corresponding to the question-and-answer prior information; wherein, the tracking data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user.

[0082] Figure 2 This is a specific embodiment of the multi-task recommendation method that integrates prior information, as described in this application.

[0083] In one embodiment, see Figure 2 The step of generating expert representation vectors for each expert and user representation vectors for the current questioning user based on the embedded data corresponding to the prior information of the question and answer includes:

[0084] S201: Input the full set of user question data and full set of expert answer data into a long short-term memory artificial neural network to obtain full set of question word vectors and full set of answer word vectors;

[0085] S202: Construct the user representation vector based on the question word vector corresponding to the current questioning user in the full set of question word vectors;

[0086] S203: Construct the expert representation vector based on the answer word vectors corresponding to each expert in the full set of answer word vectors.

[0087] Understandably, artificial neural networks can be used to construct a set of problems Q. c The corresponding full set of question word vectors and answer set A e The corresponding full-answer word vectors are used to capture the common features of all questions asked or answered by the same user or expert throughout history. The specific process is as follows:

[0088] First, an artificial neural network is used as the encoder to encode all q and a in the dataset, denoted by q and a respectively. To achieve better encoding results, the artificial neural network can be pre-trained. The original training data comes from historical data accumulated on a Q&A online communication platform; the training method follows existing neural network training methods. The artificial neural network includes at least a Long Short-Term Memory (LSTM) artificial neural network and a Natural Language Processing (NLP) framework (Bidirectional Encoder Representation from Transformers, BERT).

[0089] Second, based on historical tracking data, the questions and answers are organized into Q. c With A e Q in the following text c With A e It refers to the set of representation vectors of questions and answers.

[0090] Third, the user representation and expert representation are calculated using the following formula:

[0091]

[0092]

[0093] Where, p c With p e These are the user representation vector and the expert representation vector, respectively, |Q c | and | Q e | indicates the size of the question set and the answer set.

[0094] As can be seen from the above description, the multi-task recommendation method that integrates prior information provided in this application can generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the embedded data corresponding to the question-and-answer prior information.

[0095] Figure 3 This is a specific embodiment of the multi-task recommendation method that integrates prior information, as described in this application.

[0096] In one embodiment, see Figure 3 The step of determining multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector includes:

[0097] S301: Calculate the similarity between the expert representation vector and the user representation vector corresponding to each expert to obtain the first similarity value;

[0098] S302: Normalize and sort the first similarity value to obtain multiple experts corresponding to the current question information.

[0099] Understandably, based on expert representation vectors and user representation vectors, experts are recommended first. Once candidate experts are determined, the range of candidate answers is narrowed down (the optimal answer will only come from the set of candidate expert answers).

[0100] Specifically, based on the data points, the correspondence between q and a is obtained. A question typically corresponds to n answers, which can be represented as [a1, a2, ..., a...]. n Based on the information corresponding to the question and answer, and the information corresponding to the question user and the answer expert, obtain the user expert correspondence information, that is, the list of experts most likely to answer user c's question [e1, e2, ..., e...]. m In this process, user c asks question q, which includes candidate answer a1. A1 is answered by expert e. For user c, expert e's weight count is incremented by 1. Finally, the experts are sorted in descending order of their weight counts to obtain a list of experts. Based on a set threshold, the top K experts are selected as the multiple experts corresponding to the current question, meaning the experts who can answer the question posed by the user.

[0101] The aforementioned steps have already generated p c With p e Next, the expert recommendation task is abstracted into an information retrieval task. The similarity between the user representation vector and all expert representation vectors is calculated, where the [e1, e2, ..., e] calculated in the previous steps... m This is the baseline truth (also known as ground truth).

[0102] s(p c p e ) = p c ·p e (3)

[0103]

[0104] In the above formula, the meanings of each symbol are as follows:

[0105] In equation (3), p c p represents the user e For the expert representation corresponding to this user, s(p c p e ) represents the similarity between a user and their corresponding expert, expressed as a vector dot product.

[0106] In equation (4), x represents a specific user, x∈[0,1,2,...,n], n is the maximum number of users in the system minus 1, and i is an expert under user x, i∈[0,1,2,...,m], m is the number of experts under user x. y′ xi Let be the activation function, representing the activation function value of user x and its corresponding expert i.

[0107] The purpose of activation functions is to make the overall model's training results smoother.

[0108] Formula (3) corresponds to step S301, and formula (4) corresponds to step S302.

[0109] As can be seen from the above description, the multi-task recommendation method that integrates prior information provided in this application can determine multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector.

[0110] Figure 4 This is a specific embodiment of the multi-task recommendation method that integrates prior information, as described in this application.

[0111] In one embodiment, see Figure 4 The step of recommending the optimal answer corresponding to the current question information from all the answers provided by the multiple experts includes:

[0112] S401: Select the answer word vectors corresponding to all the answers provided by the multiple experts from the full set of answer word vectors;

[0113] S402: Calculate the similarity between each answer word vector and the question word vector corresponding to the current questioning user, and obtain the second similarity value;

[0114] S403: Select the answer corresponding to the one with the largest similarity value in the second similarity list as the optimal answer.

[0115] Understandably, following the steps described above, by setting a threshold, we can obtain the k experts most likely to answer the question. The answers from these k experts constitute the final set of candidate answers (if the baseline truth answer is not included, it can be directly added, i.e., the union of the sets is taken).

[0116] Next, a new information retrieval task is constructed by combining question q with the set of answer candidates. The similarity between question q and each answer in the set of answer candidates is calculated to determine the optimal answer.

[0117] S(p q p a ) = p q ·p a (5)

[0118]

[0119] Formulas (5) and (6) are the same as formulas (3) and (4), where p q p represents the problem representation. a The answer is represented by formula (5), which is the similarity formula, and formula (6) is the activation function formula. This is also to ensure the smoothness of the values ​​during model training.

[0120] As can be seen from the above description, the multi-task recommendation method that integrates prior information provided in this application can recommend the optimal answer corresponding to the current question information from all the answers provided by the multiple experts.

[0121] Figure 5 This is a specific embodiment of the multi-task recommendation method that integrates prior information, as described in this application.

[0122] In one embodiment, see Figure 5 After recommending the optimal answer corresponding to the current question information from all the answers provided by the multiple experts, the method further includes:

[0123] S501: Construct a recommendation error loss function based on the first similarity value and the second similarity value; specifically, see Figure 6 Calculate the expert recommendation error loss function based on the first similarity value (S601); calculate the answer recommendation error loss function based on the second similarity value (S602); construct the recommendation error loss function based on the expert recommendation error loss function and the answer recommendation error loss function (S603).

[0124] S502: Adjust the parameters of the long short-term memory artificial neural network according to the recommended error loss function.

[0125] It is understandable that the parameters of an artificial neural network can be adjusted using the cross-entropy loss function. In the embodiments of this application, adjusting the parameters of the artificial neural network allows for better representation of both the user representation vector and the expert representation vector.

[0126] In this embodiment of the application, the overall model loss is L. rec With L ir The weighted sum of the two parts. Where α is the weight.

[0127] L = L rec +αL ir (7)

[0128]

[0129]

[0130] In the formula, L rec L is the user expert loss function. ir Let be the loss function for the answer to the question.

[0131] Adaptive algorithms, such as mini-batch gradient descent or the Adam optimizer, are used in conjunction with a loss function. The loss function is the objective function that the adaptive algorithm optimizes, and its value reflects the difference between the model's predictions and the actual results. Specifically, when the model's parameters are adjusted, the value of the loss function changes accordingly. The optimization algorithm iterates and continuously adjusts the model parameters to reduce the value of the loss function until convergence is achieved. This allows for the optimization of the parameters of the artificial neural network, resulting in a more accurate vector representation. In other words, adjusting the parameters of the artificial neural network according to the recommendation error loss function includes: calculating the loss value corresponding to the artificial neural network using the recommendation error loss function; and adjusting the parameters of the artificial neural network according to the loss value until the recommendation error loss function reaches convergence.

[0132] As can be seen from the above description, the multi-task recommendation method that integrates prior information provided in this application can adjust the parameters of the long short-term memory artificial neural network according to the recommendation error loss function.

[0133] To better illustrate the technical solution of the present invention, a complete example is given here.

[0134] The first step is to deduce the correspondence between users and experts based on the correspondence between questions and answers, since questions are related to users (questioners) and answers are related to experts (responders). The text data of users and experts, as well as the text data of questions and answers, are encoded using an encoder and transformed into one-dimensional vectors C, E and Q, A. For example, c1 = [1, 0.24, ..., -0.33, -1] represents user 1, e1 = [1, 0.24, ..., -0.33, -1] represents expert 1, q1 = [1, 0.24, ..., -0.33, -1] represents question 1, and a1 = [1, 0.24, ..., -0.33, -1] represents answer 1. Since user c1 ∈ [e1, e3... e8], it means that experts e1, e3, etc., have answered this user's questions, and therefore, the experts most likely to answer this user's questions in the future are also e1, e3, etc.; c1 ∈ [q1, q3... q8] means that this user has posted questions 1, 3, etc.

[0135] The second step is to recalculate the representation vectors of the user and their corresponding expert at this moment according to formulas (1) and (2) when a user X posts a new question q. and

[0136] The third step is to calculate the similarity value between the user and the corresponding expert using formula (3), and to calculate the recommendation degree of all experts using formula (4).

[0137] The fourth step involves manually setting the recommendation score to 0.85 or higher (the parameter can be adjusted) to identify the experts who are most likely to answer the question. The answers from these k experts constitute the final set of candidate answers.

[0138] The fifth step is to calculate the similarity value between the question and the alternative answers using formula (5), and to calculate the recommendation degree of all answers using formula (6).

[0139] The sixth step involves manually setting the recommendation score to 0.98 or higher (the parameter can be adjusted) as a recommended answer and sending it to the user who asked the question.

[0140] Based on the same inventive concept, this application also provides a multi-task recommendation device that integrates prior information, which can be used to implement the method described in the above embodiments, as described in the following embodiments. Since the principle of the multi-task recommendation device integrating prior information is similar to that of the multi-task recommendation method integrating prior information, the implementation of the multi-task recommendation device integrating prior information can refer to the implementation of the method based on software performance benchmarks, and will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0141] In one embodiment, see Figure 7 In order to integrate pre-stored question-and-answer prior information for multi-task recommendation and leverage the strong correlation between data to help users quickly retrieve the optimal answer to their questions, this application provides a multi-task recommendation device that integrates prior information, including: a representation vector generation unit 701, an expert recommendation unit 702, and an answer recommendation unit 703.

[0142] The representation vector generation unit 701 is used to generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the embedded data corresponding to the question-and-answer prior information; wherein, the embedded data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user;

[0143] The expert recommendation unit 702 is used to determine multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector;

[0144] The answer recommendation unit 703 is used to recommend the optimal answer corresponding to the current question information from all the answers provided by the multiple experts.

[0145] In one embodiment, see Figure 8 The representation vector generation unit 701 includes: a full word vector generation module 801, a user representation vector generation module 802, and an expert representation vector generation module 803.

[0146] The full-volume word vector generation module 801 is used to input the full-volume user question data and full-volume expert answer data into the long short-term memory artificial neural network to obtain full-volume question word vectors and full-volume answer word vectors;

[0147] User representation vector generation module 802 is used to construct the user representation vector based on the question word vector corresponding to the current questioning user in the full set of question word vectors;

[0148] The expert representation vector generation module 803 is used to construct the expert representation vector based on the answer word vectors corresponding to each expert in the full set of answer word vectors.

[0149] In one embodiment, see Figure 9 The expert recommendation unit 702 includes: a first similarity value calculation module 901 and an expert recommendation module 902.

[0150] The first similarity value calculation module 901 is used to calculate the similarity between the expert representation vector and the user representation vector corresponding to each expert, and obtain the first similarity value.

[0151] The expert recommendation module 902 is used to normalize and sort the first similarity value to obtain multiple experts corresponding to the current question information.

[0152] In one embodiment, see Figure 10 The answer recommendation unit 703 includes: an answer word vector selection module 1001, a second similarity value calculation module 1002, and an answer recommendation module 1003.

[0153] The answer word vector selection module 1001 is used to select the answer word vectors corresponding to all the answers provided by the multiple experts from the full set of answer word vectors;

[0154] The second similarity value calculation module 1002 is used to calculate the similarity between each answer word vector and the question word vector corresponding to the current questioning user, and obtain the second similarity value;

[0155] The answer recommendation module 1003 is used to select the answer corresponding to the largest of the second similarity values ​​as the optimal answer.

[0156] In one embodiment, see Figure 11 The multi-task recommendation device that integrates prior information also includes: a recommendation loss function construction unit 1101 and a network parameter adjustment unit 1102.

[0157] The recommendation loss function construction unit 1101 is used to construct a recommendation error loss function based on the first similarity value and the second similarity value;

[0158] The network parameter adjustment unit 1102 is used to adjust the parameters of the long short-term memory artificial neural network according to the recommendation error loss function.

[0159] In one embodiment, see Figure 12 The recommendation loss function construction unit 1101 includes: an expert loss function construction module 1201, an answer loss function construction module 1202, and a recommendation loss function construction module 1203.

[0160] The expert loss function construction module 1201 is used to calculate the expert recommendation error loss function based on the first similarity value;

[0161] Answer loss function construction module 1202 is used to calculate the answer recommendation error loss function based on the second similarity value;

[0162] The recommendation loss function construction module 1203 is used to construct a recommendation error loss function based on the expert recommendation error loss function and the answer recommendation error loss function.

[0163] From a hardware perspective, in order to integrate pre-stored question-and-answer prior information for multi-task recommendation and leverage the strong correlation between data to help users quickly retrieve the optimal answer to their questions, this application provides an embodiment of an electronic device for implementing all or part of the multi-task recommendation method that integrates prior information. The electronic device specifically includes the following:

[0164] The system comprises a processor, a memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the multi-task recommendation device that integrates prior information and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the multi-task recommendation method that integrates prior information and the embodiments of the multi-task recommendation device that integrates prior information, the content of which is incorporated herein, and repeated details will not be described again.

[0165] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0166] In practical applications, the multi-task recommendation method that integrates prior information can be partially executed on the electronic device side as described above, or all operations can be completed on the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed on the client device, the client device may further include a processor.

[0167] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0168] Figure 13 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 13 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 13 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0169] In one embodiment, the multi-task recommendation method functionality that integrates prior information can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0170] S101: Generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the tracking data corresponding to the question-and-answer prior information; wherein, the tracking data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user;

[0171] S102: Determine multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector;

[0172] S103: Recommend the optimal answer corresponding to the current question information from all the answers provided by the multiple experts.

[0173] As can be seen from the above description, the multi-task recommendation method and apparatus that integrates prior information provided in this application can capture the strong correlation between questions and answers through a prototype network, and jointly train the two tasks of information retrieval and recommendation expert, targeting application scenarios and business characteristics. In the process of recommending experts, the candidate answer set is narrowed down, and finally the optimal answer to the question asked by the user is retrieved quickly, which improves the problem-solving rate of users and has achieved good results in practical applications.

[0174] In another embodiment, the multi-task recommendation device that integrates prior information can be configured separately from the central processing unit 9100. For example, the multi-task recommendation device that integrates prior information can be configured as a chip connected to the central processing unit 9100, and the function of the multi-task recommendation method that integrates prior information can be realized through the control of the central processing unit.

[0175] like Figure 13 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 13 All components shown; in addition, the electronic device 9600 may also include Figure 13 For components not shown, please refer to existing technologies.

[0176] like Figure 13 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0177] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0178] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0179] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0180] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0181] The communication module 9110 is a transmitter / receiver 9110 that transmits and receives signals via the antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which can be the same as in a conventional mobile communication terminal.

[0182] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0183] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the multi-task recommendation method for fusing prior information, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the multi-task recommendation method for fusing prior information, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0184] S101: Generate expert representation vectors for each expert and user representation vectors for the current questioning user based on the tracking data corresponding to the question-and-answer prior information; wherein, the tracking data includes all user question data and all expert answer data; the question-and-answer prior information includes the current question information of the current questioning user;

[0185] S102: Determine multiple experts corresponding to the current question information based on the user representation vector and the expert representation vector;

[0186] S103: Recommend the optimal answer corresponding to the current question information from all the answers provided by the multiple experts.

[0187] As can be seen from the above description, the multi-task recommendation method and apparatus that integrates prior information provided in this application can capture the strong correlation between questions and answers through a prototype network, and jointly train the two tasks of information retrieval and recommendation expert, targeting application scenarios and business characteristics. In the process of recommending experts, the candidate answer set is narrowed down, and finally the optimal answer to the question asked by the user is retrieved quickly, which improves the problem-solving rate of users and has achieved good results in practical applications.

[0188] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0189] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0190] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0191] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0192] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A multi-task recommendation method that integrates prior information, characterized in that, include: When the current user posts the current question, the current question and the current user are added to the question-and-answer prior information. The embedded data corresponding to the question-and-answer prior information is input into the artificial neural network to obtain the full set of question word vectors and the full set of answer word vectors; wherein, the embedded data includes the full set of user question data and the full set of expert answer data; the question-and-answer prior information includes the current question information of the current questioning user; A user representation vector is constructed based on the question word vector corresponding to the current questioner in the full set of question word vectors; the user representation vector is used to represent all the questions that the user has asked in history; An expert representation vector is constructed based on the answer word vectors corresponding to each expert in the full set of answer word vectors; the expert representation vector is used to represent all the questions that the expert has answered in history; Calculate the similarity between the expert representation vector and the user representation vector for each expert to obtain the first similarity value; The first similarity value is normalized and sorted to obtain multiple experts corresponding to the current question information; Select the answer word vectors corresponding to all the answers provided by the multiple experts from the full set of answer word vectors; Calculate the similarity between each answer word vector and the question word vector corresponding to the current questioning user to obtain a second similarity value; The answer corresponding to the largest of the second similarity values ​​is selected as the optimal answer.

2. The multi-task recommendation method integrating prior information according to claim 1, characterized in that, After recommending the optimal answer corresponding to the current question from all the answers provided by the multiple experts, the process also includes: A recommendation error loss function is constructed based on the first similarity value and the second similarity value; The parameters of the artificial neural network are adjusted according to the recommended error loss function.

3. The multi-task recommendation method integrating prior information according to claim 2, characterized in that, The step of constructing a recommendation error loss function based on the normalized first similarity value and the second similarity value includes: Calculate the expert recommendation error loss function based on the first similarity value; The answer recommendation error loss function is calculated based on the second similarity value; Construct a recommendation error loss function based on the expert recommendation error loss function and the answer recommendation error loss function.

4. The multi-task recommendation method integrating prior information according to claim 3, characterized in that, The step of adjusting the parameters of the artificial neural network according to the recommendation error loss function includes: The loss value corresponding to the artificial neural network is calculated using the recommendation error loss function; The parameters of the artificial neural network are adjusted based on the loss value until the recommendation error loss function reaches convergence.

5. A multi-task recommendation device that integrates prior information, characterized in that, include: The representation vector generation unit is used to add the current question information and the current question user to the question-answering prior information when the current question user publishes the current question information; The system inputs the embedded data corresponding to the question-and-answer prior information into the artificial neural network to obtain full question word vectors and full answer word vectors. The embedded data includes full user question data and full expert answer data. The question-and-answer prior information includes the current question information of the current questioning user. A user representation vector is constructed based on the question word vector corresponding to the current questioning user in the full question word vectors. The user representation vector represents all questions asked by the user in history. An expert representation vector is constructed based on the answer word vectors corresponding to each expert in the full answer word vectors. The expert representation vector represents all questions answered by the expert in history. An expert recommendation unit is used to calculate the similarity between the expert representation vector and the user representation vector corresponding to each expert, and obtain a first similarity value; the first similarity value is normalized and sorted to obtain multiple experts corresponding to the current question information; The answer recommendation unit is used to select the answer word vectors corresponding to all answers provided by the multiple experts from the full set of answer word vectors; calculate the similarity between each answer word vector and the question word vector corresponding to the current questioning user to obtain a second similarity value; and select the answer corresponding to the largest of the second similarity values ​​as the optimal answer.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-task recommendation method that integrates prior information as described in any one of claims 1 to 4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the multi-task recommendation method that integrates prior information as described in any one of claims 1 to 4.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the multi-task recommendation method that integrates prior information as described in any one of claims 1 to 4.

Images