A cross-domain recommendation method and system for question answering experts in question answering communities

Abstract

The present invention belongs to the field of data mining technology, and provides a method and system for cross-domain recommendation of question answering experts for question answering communities, the method extracts question and expert feature information in a target domain and a source domain; constructs training samples based on the question and expert feature information; trains a target domain expert recommendation model and a source domain expert recommendation model based on the training samples; performs expert recommendation on new questions in the target domain based on the trained target domain expert recommendation model and the source domain expert recommendation model to obtain an expert recommendation list, and helps accumulate information to enrich the community to improve the cross-domain recommendation model for recommendation performance in the new question answering community, thereby solving the problem of information scarcity in the new community and effectively improving the recommendation performance in the target domain.

Description

Technical Field

[0001] The present invention belongs to the technical field of data mining, and in particular relates to a method and system for cross-domain recommendation of question answering experts for a question-and-answer community. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Recently, with the development of the Internet, knowledge question and answer communities have increasingly become new platforms for people to share experiences and acquire knowledge. The model of knowledge question and answer communities is interactive, and a large number of users raise their own difficult questions on the platform and share their knowledge and insights. For users, they are not only questioners, but also answerers. As questioners, users can ask questions to other users on the platform to get their answers; as answerers, users can put forward their own insights based on the experience accumulated in their lives and the knowledge obtained through various channels to help users with questions solve various difficult problems. However, with the increase in questions, many problems cannot be assigned to appropriate experts in time and cannot be solved in a timely and effective manner. In order to solve this problem, expert recommendation in knowledge question and answer communities was proposed, and its main purpose is to assign appropriate experts to questions in a timely manner, so that questions can be answered quickly and with high quality.

[0004] The inventors found that previous expert recommendation methods often use information from a single question-and-answer community to recommend experts. However, for some newly-used online question-and-answer communities, there are fewer registered users in the community and filling in user personal information is not mandatory, so the user's personal description is not detailed enough, and the user's professional ability characteristics cannot be fully captured, which brings great resistance and difficulties to the recommendation of questions. At the same time, the number of question-and-answer records in newly-used online question-and-answer communities is limited, and limited data will also limit the accuracy of the recommendation model. Previous methods often train recommendation models in a single question-and-answer community, which cannot meet the requirements of fast and accurate recommendations of community expert recommendation systems for new communities. Summary of the invention

[0005] In order to solve at least one technical problem existing in the above-mentioned background technology, the present invention provides a cross-domain recommendation method and system for question answering experts for question answering communities, which makes full use of the rich information of related platforms to improve the performance of the expert recommendation system of the new platform.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A first aspect of the present invention provides a cross-domain recommendation method for question answering experts in a question answering community, comprising the following steps:

[0008] Extract question and expert feature information in the target and source domains;

[0009] Construct training samples based on question and expert feature information;

[0010] Train the target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model based on the training samples;

[0011] According to the trained target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model, experts recommend new questions in the target domain and obtain an expert recommendation list, which includes:

[0012] Perform expert recommendation based on new questions in the target domain and the target domain expert recommendation model to obtain a first expert recommendation list;

[0013] Combined with the feature dimension reduction model, the similarity calculation formula is used to find multiple similar problems corresponding to this problem in the source domain, and the source domain expert recommendation model is used to make expert recommendations for multiple similar problems to obtain a second expert recommendation list;

[0014] Based on the similarity calculation formula, a third expert recommendation list in the target domain that is similar to the second expert recommendation list is calculated; based on the merging rule, the first expert recommendation list and the third expert recommendation list are merged to obtain a final expert recommendation list.

[0015] A second aspect of the present invention provides a question answering expert cross-domain recommendation system for a question answering community, comprising:

[0016] Feature extraction module, used to extract question and expert feature information in the target domain and source domain;

[0017] A training sample construction module is used to construct training samples based on questions and expert feature information;

[0018] A training module, used to train the target domain expert recommendation model, the source domain expert recommendation model and the feature dimension reduction model based on training samples;

[0019] The expert recommendation module is used to make expert recommendations for new questions in the target domain based on the trained target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model, and obtain an expert recommendation list, which specifically includes:

[0020] Perform expert recommendation based on new questions in the target domain and the target domain expert recommendation model to obtain a first expert recommendation list;

[0021] Combined with the feature dimension reduction model, the similarity calculation formula is used to find multiple similar problems corresponding to this problem in the source domain, and the source domain expert recommendation model is used to make expert recommendations for multiple similar problems to obtain a second expert recommendation list;

[0022] Based on the similarity calculation formula, a third expert recommendation list in the target domain that is similar to the second expert recommendation list is calculated; based on the merging rule, the first expert recommendation list and the third expert recommendation list are merged to obtain a final expert recommendation list.

[0023] A third aspect of the present invention provides a computer-readable storage medium.

[0024] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the above-mentioned method for cross-domain recommendation of question answering experts for a question-answering community.

[0025] A fourth aspect of the present invention provides a computer device.

[0026] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the above-mentioned method for cross-domain recommendation of question answering experts for a question-and-answer community are implemented.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention uses a cross-domain approach to enrich the information in the information accumulation enriched community (source domain) to improve the recommendation performance in the new question-and-answer community (target domain), thereby solving the problem of information scarcity in the new community and effectively improving the recommendation performance in the target domain.

[0029] The present invention proposes a feature dimension alignment model based on an autoencoder, which solves the problem that similarity cannot be calculated due to inconsistent feature dimensions of questions in two domains and inconsistent feature dimensions of question answering experts.

[0030] The present invention obtains two recommendation lists by combining the target domain recommendation model with the source domain recommendation model, and designs a good recommendation list merging strategy, which fully utilizes the rich information in the source domain and improves the recommendation performance of the target domain.

[0031] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0033] Figure 1 It is a flowchart of a method for cross-domain recommendation of question answering experts for a question answering community according to an embodiment of the present invention;

[0034] Figure 2 It is a schematic diagram of the multi-layer BP network structure of an embodiment of the present invention;

[0035] Figure 3 2 is a schematic diagram of layer-by-layer training of a stacked denoising autoencoder according to an embodiment of the present invention;

[0036] Figure 4 It is a schematic diagram of fine-tuning the stacked denoising autoencoder according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0038] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0039] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0040] Embodiment 1

[0041] like Figure 1 As shown, this embodiment provides a cross-domain recommendation method for question answering experts in a question answering community, including the following steps:

[0042] Step 1: Extract question and expert feature information in the target domain and source domain;

[0043] As one or more embodiments, in step 1, the feature information of the target domain question includes: text features: "title", "question description" and "topic", discrete features: "question category", and numerical features: "release time" and "bounty".

[0044] The expert's feature information includes: text features: "user name" and "self-introduction", discrete features: "gender", "occupation" and "personal technical expertise", and numerical features: "wealth value", "number of replies", "number of topics" and "horse hoof gold".

[0045] The feature information of the questions in the source domain includes: text features: "title", "question description" and "topic", discrete features: "question category", numerical features: "release time", "completion time", "view count", "collection count", "reward", "follow count" and "comment count";

[0046] The expert text features are: "user name" and "self-introduction", the discrete features are: "gender", "occupation" and "personal technical expertise", and the numerical features are: "wealth value", "age" and "number of topics".

[0047] It can be understood that the target domain and the source domain can be selected and set according to actual needs. For example, in this embodiment, the target domain can be the Houpao Forum and the source domain can be the Haichuan Chemical Forum.

[0048] Step 2: Construct training samples based on question and expert feature information;

[0049] As one or more embodiments, in step 2, constructing training samples based on the question and expert feature information specifically includes:

[0050] Step 201: constructing a scoring matrix of the target domain and the auxiliary domain based on the interactive information between the questions in the target domain and the source domain and the experts;

[0051] Assume that the expert sets of the target domain and the source domain are U T and U S , the lengths of the two are m T and m S , the set of questions in the target domain and the source domain is Q T and Q S , the lengths are n T and n S , the set of cold start problems in the target domain Q New .

[0052] The number of likes obtained by the expert in answering the question in the target domain is used as the score, and the score is normalized to [0,5] using formula (1) to obtain the target domain score matrix The normalization formula is:

[0053]

[0054] Among them, x max is the maximum number of likes among all answers to a question, xmin is the minimum value.

[0055] The wealth value obtained by the expert answering the question in the source domain is used as the score, and the score is processed in the same normalization method as the target domain. The final source domain score matrix is:

[0056] Step 202: Use the natural language processing model Bert model to convert the text features of the question and the expert into vector representations, use one-hot representation for the discretized features, and normalize the numerical features. Combine the three parts of features to form a vectorized feature representation of the question and the expert.

[0057] For a question or an expert's text feature we synthesize all fields into a description text.

[0058] Experts in the target domain The text features are converted into embedded vectors using the natural language processing model Bert, discrete features are encoded using one-hot, and numerical features are normalized using the following formula:

[0059]

[0060] Among them, min is the maximum value of the numerical feature and max is the minimum value.

[0061] After combining the above three features, we get the expert Vectorized features of Similarly, the problem in the target domain The vectorized representation of

[0062] Experts in the source domain The text features are converted into embedding vectors using the natural language processing model Bert, and then concatenated with the one-hot encoded discrete vector and the normalized numerical vector to obtain the expert Vectorized features of Similarly, the problem in the target domain The vectorized representation of

[0063] Step 203: construct training sample data based on the vectorized feature representation and rating matrix of the questions and experts in the target domain and the source domain;

[0064] After the vectorization of experts and questions in the two fields, the data for training the expert recommendation model can be constructed according to the scores in the score matrix of the target domain and the auxiliary domain. Let the score of question q corresponding to expert u be r uq , then we can construct training samples in is feature concatenation, r uq The score corresponding to this sample. In this way, the training samples of the target domain and the auxiliary domain can be constructed respectively.

[0065] Step 3: Train the target domain expert recommendation model and the source domain expert recommendation model based on the training samples;

[0066] The BP network has good fitting ability, so this embodiment uses a multi-layer BP network to perform expert recommendation score prediction.

[0067] The structure diagram of the multi-layer BP network with L layers is as follows Figure 2 As shown, the calculation process is as follows:

[0068]

[0069] h (1) =relu(h (0) W (1) +b (1) )

[0070] h (2) =relu(h (1) W (2) +b (2) ) ...

[0072] h (L) =relu(h (L-1) W (L) +b (L) )

[0073]

[0074] Among them, relu is a nonlinear activation function, W and b are weight matrices and biases.

[0075] The optimization objectives of the rating prediction network are as follows:

[0076]

[0077] Where D is the data set, λ is the regularization coefficient, θ={W,b} is the parameter set, ||·|| is the F-norm of the matrix, is the score between expert u and question q predicted by the model.

[0078] This embodiment uses Adam to optimize model parameters. After the model training is completed, for a new question, its vectorized representation is concatenated with the vectorized representation of existing experts and then input into the network prediction score, and the N experts with the highest prediction scores are recommended for this question.

[0079] The above model can be used to train the expert recommendation models RecModel_T(q) and RecModel_S(q) for the target domain and source domain respectively. The input of this recommendation model is the feature vector representation of the question and the expert, and the output is an expert recommendation list of length N.

[0080] Step 4: Train the feature dimension reduction model of the target domain and auxiliary domain;

[0081] This embodiment uses similarity calculation to find similar problems in the source domain for new problems in the target domain, recommends experts for similar problems with the help of the auxiliary domain recommendation model, and finally determines the expert recommendation list of the target domain by calculating the similarity between the experts in the auxiliary domain recommended expert list and the experts in the target domain. However, the problem feature dimensions of the target domain and the source domain are inconsistent, and the expert feature dimensions are also inconsistent, which makes it impossible to perform similarity calculation.

[0082] This embodiment will use a feature dimensionality reduction model based on stacked denoising autoencoders (SDAE) to map the problem features of the two fields to the same dimension, and do the same for the expert features. SDAE is composed of multiple denoising autoencoders stacked together, and a single denoising autoencoder consists of an encoder and a decoder. The calculation formula of the encoder is as follows:

[0083]

[0084] in, The feature after adding Gaussian noise to the encoder input x, W Encoder ,b Encoder are the weight and bias of the encoder respectively, and LeakyRelu is the activation function.

[0085] Similarly, the decoder calculation formula is as follows:

[0086] z=LeakyRelu(W Decoder y+b Decoder )

[0087] Among them, W Decoder ,b Decoder are the weights and biases of the decoder, LeakyRelu is the activation function, and z is the reconstructed representation of the encoder input x.

[0088] The loss function of the denoising autoencoder is as follows:

[0089]

[0090] In the target domain and source domain, the existing expert features and problem features are used to train the expert feature stacked denoising autoencoder and the problem stacked denoising autoencoder in their respective fields. The training process adopts a layer-by-layer training and fine-tuning training method, and Adam is used to optimize the model parameters. The schematic diagram of layer-by-layer training is shown in the figure. Figure 3 As shown, after the layer-by-layer training is completed, all layers are connected to fine-tune the pre-trained parameters. The fine-tuning diagram is shown in Figure 4 shown.

[0091] Finally, the target domain can be obtained to map the problem features to a low-dimensional encoder UEC T (·), mapping expert features to a low-dimensional encoder QEC T (·), the source domain maps the problem features to a low-dimensional encoder UEC S (·), mapping expert features to a low-dimensional encoder QEC S (·) UEC T (·) and UEC S (·) Expert features can be mapped to the same dimension d U , QEC T (·) and QEC S (·) Problem features can be mapped to the same dimension d Q .

[0092] Step 5: Perform expert recommendation on new questions in the target domain based on the trained target domain expert recommendation model and source domain expert recommendation model to obtain an expert recommendation list;

[0093] This embodiment uses the target domain recommendation model and the auxiliary domain recommendation model to jointly recommend experts in the target domain for new questions in the target domain. A recommendation method is designed to recommend experts for new questions in a newly invested question and answer community (target domain) by leveraging a question and answer community (source domain) with rich information accumulation.

[0094] For a new question q in the target domain new , the recommendation process has the following steps:

[0095] (1) Convert the text features, discrete features, and numerical features of the new question into vector representations to obtain the feature representation of the question;

[0096] (2) Using the target domain recommendation model to make recommendations in the target domain, a first expert recommendation list list_A is obtained, and the length of the recommendation list is N;

[0097] (3) Find K problems similar to this problem in the source domain by calculating similarity, and use the recommendation model of the auxiliary domain (i.e., the source domain) to recommend KN experts for these similar problems. The top N experts are obtained to obtain the second expert list list_B;

[0098] Using Encoder QEC T (·) The new problem q in the target domain new The feature map is of dimension d Q Vector representation of Using QEC S (·) Map the features of the source domain problem into a dimension of d Q The feature vector representation of The K questions closest to the new question are found by calculating the cosine similarity between the new question and the question in the source domain. The similarity calculation formula is as follows:

[0099]

[0100] (4) By calculating the similarity, a third expert list list_C in the target domain that is similar to the experts in the second expert list list_B is obtained, and the length of the recommended list is N;

[0101] Using the encoder UEC S (·) Map the features of the experts in this recommendation list to d U Dimension, while using UEC T (·) Map the expert features in the target domain to d U Dimension, use cosine similarity calculation to find the expert in the target domain that is most similar to each expert in the second expert recommendation list list_B to obtain the third expert recommendation list list_C.

[0102] (5) Combine the first expert recommendation list list_A and list_C to obtain the final recommendation list list_D.

[0103] The expert recommendation list list_A and the third expert recommendation list list_C are merged to obtain a final expert recommendation list list_D of length N.

[0104] The merging rules are:

[0105] ① Recommend that the experts common to the first expert recommendation list list_A and the third expert recommendation list list_C be directly added to the fourth expert recommendation list list_D, and arranged in the order of the experts in the third expert recommendation list list_C.

[0106] ② If the recommended list list_D still has less than N experts after the processing of ①, half of the number of the missing experts will be selected from the third expert recommendation list list_C. If the number is not divisible, the number of experts selected from list_C will be rounded up, and the remainder will be selected from list_A. The selected experts will maintain the original arrangement order.

[0107] If list_A={u1,u3,u 10 ,u2,u8},list_C={u 10 ,u7,u1,u4,u 15}, then list_D={u 10 ,u1,u7,u4,u3}.

[0108] Embodiment 2

[0109] This embodiment provides a question answering expert cross-domain recommendation system for a question answering community, including:

[0110] Feature extraction module, used to extract question and expert feature information in the target domain and source domain;

[0111] A training sample construction module is used to construct training samples based on questions and expert feature information;

[0112] A training module, used to train the target domain expert recommendation model, the source domain expert recommendation model and the feature dimension reduction model based on training samples;

[0113] The expert recommendation module is used to make expert recommendations for new questions in the target domain based on the trained target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model, and obtain an expert recommendation list, which specifically includes:

[0114] Perform expert recommendation based on new questions in the target domain and the target domain expert recommendation model to obtain a first expert recommendation list;

[0115] Combined with the feature dimension reduction model, the similarity calculation formula is used to find multiple similar problems corresponding to this problem in the source domain, and the source domain expert recommendation model is used to make expert recommendations for multiple similar problems to obtain a second expert recommendation list;

[0116] Based on the similarity calculation formula, a third expert recommendation list in the target domain that is similar to the second expert recommendation list is calculated; based on the merging rule, the first expert recommendation list and the third expert recommendation list are merged to obtain a final expert recommendation list.

[0117] Embodiment 3

[0118] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the above-mentioned method for cross-domain recommendation of question answering experts for a question and answer community are implemented.

[0119] Embodiment 4

[0120] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the above-mentioned method for cross-domain recommendation of question answering experts for a question-and-answer community are implemented.

[0121] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0122] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0123] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0124] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0125] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A cross-domain recommendation method for question answering experts in a question answering community, characterized in that: The steps include: Extract question and expert feature information in the target and source domains; Construct training samples based on question and expert feature information; Train the target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model based on the training samples; According to the trained target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model, experts recommend new questions in the target domain and obtain an expert recommendation list, which includes: Perform expert recommendation based on new questions in the target domain and the target domain expert recommendation model to obtain a first expert recommendation list; Combined with the feature dimension reduction model, the similarity calculation formula is used to find multiple similar problems corresponding to this problem in the source domain, and the source domain expert recommendation model is used to make expert recommendations for multiple similar problems to obtain a second expert recommendation list; Based on the similarity calculation formula, a third expert recommendation list similar to the second expert recommendation list in the target domain is calculated; based on the merging rule, the first expert recommendation list and the third expert recommendation list are merged to obtain a final expert recommendation list.

2. A cross-domain recommendation method for question answering experts for a question answering community as claimed in claim 1, characterized in that: The constructing of training samples based on question and expert feature information includes: Construct a scoring matrix for the target domain and the auxiliary domain based on the interactive information between the questions and experts in the target domain and the source domain; The natural language processing model Bert model is used to convert the text features of questions and experts into vector representations. Discrete features are represented using one-hot representations, and numerical features are normalized. The corresponding features are combined to form vectorized feature representations of questions and experts. The training sample data is constructed based on the vectorized feature representation and rating matrix of questions and experts in the target domain and source domain.

3. The cross-domain recommendation method for question answering experts for a question answering community as claimed in claim 1, characterized in that: The similarity calculation formula is used to find multiple similar questions corresponding to this question in the source domain, and the source domain expert recommendation model is used to perform expert recommendation on the multiple similar questions to obtain a second expert recommendation list, including: Use the target domain encoder to map the features of the new problem into a vector representation At the same time, the source domain encoder is used to map the features of the source domain problem into a vector representation The K questions that are most similar to the new question are found by calculating the cosine similarity between the features of the new question and the questions in the source domain.

4. The cross-domain recommendation method for question answering experts for a question answering community as claimed in claim 1, characterized in that: The feature dimension reduction model construction process includes: using stacked denoising autoencoders in the target domain and the source domain respectively to map the problem feature representation and the expert feature representation in the two domains to the same feature dimension.

5. The cross-domain recommendation method for question answering experts for a question answering community as claimed in claim 4, characterized in that: The stacked denoising autoencoder includes multiple denoising autoencoders, and a single denoising autoencoder consists of an encoder and a decoder. The calculation formula of the encoder is as follows: in, The feature after adding Gaussian noise to the encoder input x, W Encoder ,b Encoder are the weight and bias of the encoder respectively, and LeakyRelu is the activation function; The calculation formula of the decoder is as follows: z=LeakyRelu(W Decoder y+b Decoder ) Among them, W Decoder ,b Decoder are the weights and biases of the decoder, LeakyRelu is the activation function, and z is the reconstructed representation of the encoder input x.

6. The cross-domain recommendation method for question answering experts for a question answering community as claimed in claim 1, characterized in that: The target domain expert recommendation model and the source domain expert recommendation model adopt a score prediction network based on a multi-layer BP network, and the optimization objective function of the score prediction network is: Where D is the data set, λ is the regularization coefficient, θ={W,b} is the parameter set, ||·|| is the F-norm of the matrix, r uq is the score of question v corresponding to expert u, is the score between expert u and question q predicted by the model.

7. The cross-domain recommendation method for question answering experts for a question answering community as claimed in claim 1, characterized in that: The merging rules are: If there are common experts in the first expert recommendation list and the third expert recommendation list, they are directly added to the fourth expert recommendation list and arranged in the order of the third expert recommendation list; If the set number of experts is still insufficient after the fourth expert recommendation list, half of the total number of missing experts will be selected from the third recommendation list. If the number is not divisible, the number of experts selected from the third recommendation list will be rounded up, and the remaining experts will be selected from the first expert recommendation list, and the selected experts will maintain the original arrangement order.

8. A question answering expert cross-domain recommendation system for question answering communities, characterized in that: include: Feature extraction module, used to extract question and expert feature information in the target domain and source domain; A training sample construction module is used to construct training samples based on questions and expert feature information; A training module, used to train the target domain expert recommendation model, the source domain expert recommendation model and the feature dimension reduction model based on training samples; The expert recommendation module is used to make expert recommendations for new questions in the target domain based on the trained target domain expert recommendation model, source domain expert recommendation model and feature dimension reduction model, and obtain an expert recommendation list, which specifically includes: Perform expert recommendation based on new questions in the target domain and the target domain expert recommendation model to obtain a first expert recommendation list; Combined with the feature dimension reduction model, the similarity calculation formula is used to find multiple similar problems corresponding to this problem in the source domain, and the source domain expert recommendation model is used to make expert recommendations for multiple similar problems to obtain a second expert recommendation list; Based on the similarity calculation formula, a third expert recommendation list similar to the second expert recommendation list in the target domain is calculated; based on the merging rule, the first expert recommendation list and the third expert recommendation list are merged to obtain a final expert recommendation list.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in a cross-domain recommendation method for question answering experts for a question-and-answer community as described in any one of claims 1 to 7 are implemented.

10. A computer 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, the steps in the cross-domain recommendation method for question answering experts for a question and answer community are implemented as described in any one of claims 1-7.

Images