CN114329184A - Expert recommendation method and system in question and answer website - Google Patents
Expert recommendation method and system in question and answer website Download PDFInfo
- Publication number
- CN114329184A CN114329184A CN202111469719.6A CN202111469719A CN114329184A CN 114329184 A CN114329184 A CN 114329184A CN 202111469719 A CN202111469719 A CN 202111469719A CN 114329184 A CN114329184 A CN 114329184A
- Authority
- CN
- China
- Prior art keywords
- answer
- question
- expert
- vector
- feature vector
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 32
- 239000013598 vector Substances 0.000 claims abstract description 170
- 238000013528 artificial neural network Methods 0.000 claims abstract description 16
- 125000004122 cyclic group Chemical group 0.000 claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims description 24
- 230000006870 function Effects 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 9
- 238000003860 storage Methods 0.000 claims description 6
- 238000010586 diagram Methods 0.000 description 11
- 238000004590 computer program Methods 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 230000009193 crawling Effects 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an expert recommendation method and system in a question and answer website, belonging to the technical field of information retrieval, wherein the method comprises the following steps: acquiring question and answer feature vectors of all users, and inputting the question and answer feature vectors into a pre-established gated cyclic network model to obtain time domain semantic feature vectors; constructing a common answer-concern network graph according to the common answer and concern relationship of the users; inputting the time domain semantic feature vector into a pre-established graph convolution neural network to learn a topological structure of a common answer-concern network graph to obtain a spatial domain social feature vector; inputting the airspace social contact feature vectors into a full-connection layer for classification to obtain classification feature vectors, wherein users corresponding to elements with the median value of 1 in the classification feature vectors are experts to obtain an expert list; the method comprises the steps of obtaining a user feature vector of an expert and a feature vector of an appointed problem, calculating cosine similarity of the user feature vector of the expert and the feature vector of the appointed problem, and taking the expert with the cosine similarity larger than a preset threshold value as a recommendation result.
Description
Technical Field
The invention relates to an expert recommendation method and system in a question and answer website, and belongs to the technical field of information retrieval.
Background
In the quiz web site such as quiz, quica, etc., finding an expert related to a given Question field and willing to answer the Question to provide a high quality answer not only eliminates the confusion of the questioner, but also attracts more users to participate in discussion and join interested communities, which makes the content of the web site more comprehensive and the questioner more easily get a satisfactory answer, which is very important for the development of cqa (community Question answering).
In the prior art, the user is usually distinguished by identifying the authentication content filled by the user, but in practical application, many users cannot fill the authentication content, the fields related to the authentication content filled by the users are not comprehensive, the accuracy is low, the coverage rate of the fields discovered by the experts in the fields is low, and the accuracy of the corresponding fields of the recommended experts meeting the problems is not high.
Disclosure of Invention
The invention aims to provide an expert recommendation method and system in a question and answer website, which improve the accuracy of expert recommendation, reduce the complexity of recommendation calculation and improve the field conformity degree of recommended experts and specified problems.
In order to realize the purpose, the invention is realized by adopting the following technical scheme:
in a first aspect, the invention provides an expert recommendation method in a question and answer website, comprising the following steps:
acquiring question and answer feature vectors of all users, and inputting the question and answer feature vectors into a pre-established gated cyclic network model to obtain time domain semantic feature vectors;
constructing a common answer-concern network graph according to the common answer and concern relationship of the users;
inputting the time domain semantic feature vector into a pre-established graph convolution neural network to learn a topological structure of a common answer-concern network graph to obtain a spatial domain social feature vector;
inputting the airspace social contact feature vectors into a full-connection layer for classification to obtain classification feature vectors, wherein users corresponding to elements with the median value of 1 in the classification feature vectors are experts to obtain an expert list;
the method comprises the steps of obtaining a user feature vector of an expert and a feature vector of an appointed problem, calculating cosine similarity of the user feature vector of the expert and the feature vector of the appointed problem, and taking the expert with the cosine similarity larger than a preset threshold value as a recommendation result.
With reference to the first aspect, further, the question-answer feature vectors of all users are obtained by:
wherein
n is the total number of users, Bert (QA)n) Embedding words obtained by performing Bert processing on the question and answer text of the nth user.
With reference to the first aspect, further, the common answer-interest network graph is constructed by:
in the question and answer website, any two users answer the same question together or one user pays attention to another user, and then one edge is connected between the two users, so that a graph G (V, E) is constructed, and the graph G (V, E) is a network graph for jointly answering and paying attention to; wherein, V is a user set, and E is an edge set.
With reference to the first aspect, further, the question-answer feature vector is input into a pre-established gated cyclic network model to obtain a time domain semantic feature vector:
ht＝ut*ht-1+(1-ut)*ct
wherein h ist-1For the output of the gated loop network model at time t-1, htThe final output of the gated cyclic network model is a time domain semantic feature vector, sigma represents a first nonlinear activation function, WuIndicating that the gate weight matrix is updated,
a question-answer feature vector representing time t, buRepresents the update gate offset vector, utTo update the gate, the degree to which the output of the previous time at the gate enters the current time, W, is updatedrRepresenting a forgetting gate weight matrix, brRepresenting a forgetting gate bias vector, rtFor the forgetting gate, the forgetting gate controls the forgetting degree output at the previous moment, tanh represents a hyperbolic tangent function, WcRepresenting a memory weight matrix, bcRepresenting a memory offset vector, ctIndicating the state stored in memory at time t.
With reference to the first aspect, further, the time domain semantic feature vector is input into a pre-established graph convolution neural network to learn a topological structure of a common answer-attention network graph, so as to obtain a spatial domain social feature vector:
H0＝L
wherein L is a time domain semantic feature vector, H0Input to a convolutional neural network, Hl+1Represents the output of the ith layer map convolutional neural network, represents a first nonlinear activation function,
a degree matrix representing a graph of the common answer-interest network,
is an adjacency matrix with self-joins added, A is an adjacency matrix of a common answer-interest network graph, INIs an identity matrix, HlRepresents the output of the l-1 th layer convolutional neural network, WlAnd a trainable weight matrix representing the l-th graph convolution neural network, wherein the output of the last layer of the graph convolution neural network is an airspace social characteristic vector.
Combining the first aspect, further, inputting the airspace social characteristic vector into the full-link layer for classification to obtain a classification characteristic vector:
YU＝softmax(δ(W2Z+b2))
where δ represents a second nonlinear activation function, W1Trainable weight matrices representing fully connected layers of the first layer, XUIs a feature vector of the user, H is a spatial social feature vector, b1Is the offset vector of the first layer fully-connected layer, Z is the output of the first layer fully-connected layer, W2Trainable weight matrices representing fully connected layers of the second layer, b2Is the offset vector of the second layer fully-connected layer, YUIs a classification feature vector.
With reference to the first aspect, the method further includes the step of marking users meeting preset requirements as experts:
if the praise number of the user in a certain week is larger than the preset praise number threshold value, the user is marked as an expert in the week, and if the user does not answer the question in a certain week, the praise number in the week is continued.
In a second aspect, the present invention further provides an expert recommendation system in a question and answer website, including:
a time domain feature extraction module: the system comprises a data processing module, a data processing module and a data processing module, wherein the data processing module is used for acquiring question and answer characteristic vectors of all users and inputting the question and answer characteristic vectors into a pre-established gating circulation network model to obtain time domain semantic characteristic vectors;
the network graph building module: the system comprises a common answer-concern network graph and a common answer-concern network graph, wherein the common answer-concern network graph is constructed according to common answers and concern relations of users;
the spatial domain feature extraction module: the system comprises a time domain semantic feature vector input unit, a public answer-attention network graph learning unit and a public domain social feature vector input unit, wherein the time domain semantic feature vector is input into a pre-established graph convolution neural network to learn a topological structure of a common answer-attention network graph, and a spatial domain social feature vector is obtained;
an expert extraction module: the system comprises a spatial domain social characteristic vector input full-connection layer, a classification characteristic vector and an expert list, wherein the classification characteristic vector is obtained by inputting the spatial domain social characteristic vector into the full-connection layer for classification, and a user corresponding to an element with a median value of 1 in the classification characteristic vector is an expert;
a recommendation module: the system is used for obtaining the user characteristic vector and the specified problem characteristic vector of the expert, calculating the cosine similarity of the user characteristic vector and the specified problem characteristic vector of the expert, and taking the expert with the cosine similarity larger than the preset threshold value as a recommendation result.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides an expert recommendation method and system in a question and answer website.A question and answer characteristic vector is input into a pre-established gated cyclic network model, the question and answer characteristic vector which dynamically changes along with time is learned by using the gated cyclic network model to capture a time domain semantic characteristic vector, and the actual situation that the knowledge storage and the professional skill of each user dynamically increase along with time is considered; a common answer-concern network graph is constructed according to the common answer and concern relationship of the user, the problem found by an expert is converted into a graph node classification problem, and the complexity of recommendation calculation is reduced; by calculating the cosine similarity of the user feature vector of the expert and the feature vector of the specified problem, the expert with the cosine similarity larger than the preset threshold is taken as a recommendation result, and the field conformity degree of the recommended expert and the specified problem is improved.
Drawings
FIG. 1 is a flowchart of an expert recommendation method in a question and answer website according to an embodiment of the present invention;
FIG. 2 is a diagram of a common answer-focus network provided by an embodiment of the present invention;
FIG. 3 is a diagram of a GRU-GCN model provided by an embodiment of the invention.
Detailed Description
The present invention is further described with reference to the accompanying drawings, and the following examples are only for clearly illustrating the technical solutions of the present invention, and should not be taken as limiting the scope of the present invention.
Example 1
As shown in fig. 1, an embodiment of the present invention provides an expert recommendation method in a question and answer website, including:
and acquiring question-answer characteristic vectors of all users, and inputting the question-answer characteristic vectors into a pre-established gated cyclic network model to obtain time domain semantic characteristic vectors.
Obtaining desensitized user information and user question and answer information from a question and answer website, storing the desensitized user information and user question and answer information in a server, preprocessing data, namely cleaning the question and answer text and removing irrelevant characters.
In the process of expert discovery, the question-answer text of a user needs to be converted into a question-answer feature vector through Bert, and the question-answer text contains partial code segments, formulas and some useless symbols, which become noise of the question-answer feature vector, so that really useful information needs to be extracted from the question-answer text, and the question-answer text is filtered by using a regular expression.
Constructing a question-answer feature vector, recording question-answer texts provided by users who are respondents every week, and if the users do not answer in a certain week, using the question-answer texts in the previous week to subject the question-answer texts of the users to Bert processing, as shown in fig. 3, the question-answer feature vector of all the users can be expressed as:
wherein
n is the total number of users, Bert (QA)n) Word embedding obtained by Bert processing for the question-answer text of the nth user, as shown in FIG. 3
To
The question-answer texts respectively representing the 1 st to 6 th users are embedded by words obtained by the Bert processing.
The method also comprises a step of marking experts according to the praise number of the user every week, if the praise number of a certain week of the user is larger than a preset praise number threshold value, the user is marked as the expert in the week, and if the user does not answer the question in a certain week, the praise number of the week is continued.
Putting the question-answer feature vectors into a pre-established gated-circular network model (GRU) to learn the question-answer feature vectors which dynamically change with time to capture time domain semantic feature vectors:
ht＝ut*ht-1+(1-ut)*ct
wherein h ist-1For the output of the gated loop network model at time t-1, htThe final output of the gated cyclic network model is a time domain semantic feature vector, sigma represents a first nonlinear activation function, WuIndicating that the gate weight matrix is updated,
a question-answer feature vector representing time t, buRepresents the update gate offset vector, utTo update the gate, the degree to which the output of the previous time at the gate enters the current time, W, is updatedrRepresenting a forgetting gate weight matrix, brRepresenting a forgetting gate bias vector, rtFor the forgetting gate, the forgetting gate controls the forgetting degree output at the previous moment, tanh represents a hyperbolic tangent function, WcRepresenting a memory weight matrix, bcRepresenting a memory offset vector, ctIndicating the state stored in memory at time t.
And constructing a common answer-concern network diagram according to the common answer and concern relationship of the users.
As shown in fig. 2, in the question and answer website, when any two users answer the same question together or a user pays attention to another user, there is an edge between the two users, and accordingly, a graph G ═ V, E > is constructed, and the graph G ═ V, E > is a network graph of joint answer-attention; wherein G is an undirected graph, V is a user set, and E is an edge set.
And inputting the time domain semantic feature vector into a pre-established graph convolution neural network to learn a topological structure of a common answer-concern network graph, so as to obtain a spatial domain social feature vector.
As shown in fig. 3, the output of the GRU is put into a pre-established graph convolution neural network (GCN) as an initial vector, an adjacency matrix with G ═ V, E > is also put into the GCN to extract information from the network, and by stacking GCN layers, the output H of the last layer of GCN is a spatial domain social feature vector, and H includes not only the own features of the node but also the features of its own neighbors.
H0＝L
Wherein L is a time domain semantic feature vector, H0Input to a convolutional neural network, Hl+1Represents the output of the ith layer map convolutional neural network, represents a first nonlinear activation function,
a degree matrix representing a graph of the common answer-interest network,
is an adjacency matrix with self-joins added, A is an adjacency matrix of a common answer-interest network graph, INIs an identity matrix, HlRepresents the output of the l-1 th layer convolutional neural network, WlAnd a trainable weight matrix representing the l-th graph convolution neural network, wherein the output of the last layer of the graph convolution neural network is an airspace social characteristic vector.
Inputting the airspace social contact feature vectors into a full-connection layer for classification to obtain classification feature vectors, and taking users corresponding to elements with the median value of 1 in the classification feature vectors as experts to obtain an expert list.
YU＝softmax(δ(W2Z+b2))
Where δ represents a second nonlinear activation function, W1Trainable weight matrices representing fully connected layers of the first layer, XUIs the feature vector of the user, X in FIG. 3U1To XU6Feature vectors representing 1 st to 6 th users, H is a spatial social feature vector, b1Is the offset vector of the first layer fully-connected layer, Z is the first layer fully-connected layerOutput of (W)2Trainable weight matrices representing fully connected layers of the second layer, b2Is the offset vector of the second layer fully-connected layer, YUIs a classification feature vector.
XUThe feature vector of the user corresponds to the user data of the user, the user data comprises five attributes, namely fan number, user introduction, whether the user is an organization user, whether the user is an advertising user or not and whether the user is an anonymous user or not, each attribute corresponds to a one-dimensional vector, and thus the length of the feature vector of the user corresponding to the user data is 5, namely XU∈Rn×5And n is the total number of users.
YUWhen the ith element is 1, the ith user is indicated as an expert, and an expert list is obtained.
The method comprises the steps of obtaining a user feature vector of an expert and a feature vector of an appointed problem, calculating cosine similarity of the user feature vector of the expert and the feature vector of the appointed problem, and taking the expert with the cosine similarity larger than a preset threshold value as a recommendation result.
When an expert needs to be recommended in case of a problem, the text of the specified problem is processed by Bert to obtain the specified problem characteristic, and the user characteristic vector X of the expert is obtainedUAnd specifying the problem feature vector, calculating the cosine similarity between the user feature vector of the expert and the specified problem feature vector, and taking the expert with the cosine similarity larger than a preset threshold value as a recommendation result, or taking the expert with the cosine similarity larger than the first n times of the preset threshold value as the recommendation result.
Example 2
The embodiment of the invention provides an expert recommendation system in a question and answer website, which comprises the following steps:
a time domain feature extraction module: the system comprises a data processing module, a data processing module and a data processing module, wherein the data processing module is used for acquiring question and answer characteristic vectors of all users and inputting the question and answer characteristic vectors into a pre-established gating circulation network model to obtain time domain semantic characteristic vectors;
the network graph building module: the system comprises a common answer-concern network graph and a common answer-concern network graph, wherein the common answer-concern network graph is constructed according to common answers and concern relations of users;
the spatial domain feature extraction module: the system comprises a time domain semantic feature vector input unit, a public answer-attention network graph learning unit and a public domain social feature vector input unit, wherein the time domain semantic feature vector is input into a pre-established graph convolution neural network to learn a topological structure of a common answer-attention network graph, and a spatial domain social feature vector is obtained;
an expert extraction module: the system comprises a spatial domain social characteristic vector input full-connection layer, a classification characteristic vector and an expert list, wherein the classification characteristic vector is obtained by inputting the spatial domain social characteristic vector into the full-connection layer for classification, and a user corresponding to an element with a median value of 1 in the classification characteristic vector is an expert;
a recommendation module: the system is used for obtaining the user characteristic vector and the specified problem characteristic vector of the expert, calculating the cosine similarity of the user characteristic vector and the specified problem characteristic vector of the expert, and taking the expert with the cosine similarity larger than the preset threshold value as a recommendation result.
Example 3
The embodiment of the invention provides an expert recommendation system in a question and answer website, which comprises the following steps:
a data extraction module: the method is used for continuously crawling required information, such as the question and answer texts of all users, from the community question and answer website by using a crawler, extracting relevant data and processing the data.
A storage module: the method and the device are used for effectively storing the crawled information, the data storage is carried out by adopting sqlite, and all database operations are realized by calling the interface.
An algorithm module: the expert recommendation method used in the question and answer website described in embodiment 1 is used for finding an expert and taking the expert corresponding to the condition that the cosine similarity between the user feature vector of the expert and the specified question feature vector is greater than the preset threshold value as a recommendation result.
A service module: for providing the functions implemented by the system to the user in the form of Web services.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may 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, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may 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, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (8)
1. An expert recommendation method in a question and answer website is characterized by comprising the following steps:
acquiring question and answer feature vectors of all users, and inputting the question and answer feature vectors into a pre-established gated cyclic network model to obtain time domain semantic feature vectors;
constructing a common answer-concern network graph according to the common answer and concern relationship of the users;
inputting the time domain semantic feature vector into a pre-established graph convolution neural network to learn a topological structure of a common answer-concern network graph to obtain a spatial domain social feature vector;
inputting the airspace social contact feature vectors into a full-connection layer for classification to obtain classification feature vectors, wherein users corresponding to elements with the median value of 1 in the classification feature vectors are experts to obtain an expert list;
the method comprises the steps of obtaining a user feature vector of an expert and a feature vector of an appointed problem, calculating cosine similarity of the user feature vector of the expert and the feature vector of the appointed problem, and taking the expert with the cosine similarity larger than a preset threshold value as a recommendation result.
2. The expert recommendation method in question-answering website according to claim 1, wherein the question-answering feature vectors of all users are obtained by the following method:
wherein
n is the total number of users, Bert (QA)n) Embedding words obtained by performing Bert processing on the question and answer text of the nth user.
3. The expert recommendation method in question-answering website according to claim 1, wherein the common answer-concern network graph is constructed by the following method:
in the question and answer website, any two users answer the same question together or one user pays attention to another user, and then one edge is connected between the two users, so that a graph G (V, E) is constructed, and the graph G (V, E) is a network graph for jointly answering and paying attention to; wherein, V is a user set, and E is an edge set.
4. The expert recommendation method in question-answering websites according to claim 1, characterized in that the question-answering feature vectors are input into a pre-established gated cyclic network model to obtain time domain semantic feature vectors:
ht＝ut*ht-1+(1-ut)*ct
wherein h ist-1For the output of the gated loop network model at time t-1, htThe final output of the gated cyclic network model is a time domain semantic feature vector, sigma represents a first nonlinear activation function, WuIndicating that the gate weight matrix is updated,
a question-answer feature vector representing time t, buRepresents the update gate offset vector, utTo update the gate, the degree to which the output of the previous time at the gate enters the current time, W, is updatedrRepresenting a forgetting gate weight matrix, brRepresenting a forgetting gate bias vector, rtFor the forgetting gate, the forgetting gate controls the forgetting degree output at the previous moment, tanh represents a hyperbolic tangent function, WcRepresenting storage weight momentsArray, bcRepresenting a memory offset vector, ctIndicating the state stored in memory at time t.
5. The expert recommendation method in question-answering websites according to claim 1, characterized in that the time domain semantic feature vectors are input into a pre-established graph convolution neural network to learn the topological structure of a common answer-concern network graph, and obtain airspace social feature vectors:
H0＝L
wherein L is a time domain semantic feature vector, H0Input to a convolutional neural network, Hl+1Representing the output of the l-th layer of the convolutional neural network, v represents a first nonlinear activation function,
a degree matrix representing a graph of the common answer-interest network,
is an adjacency matrix with self-joins added, A is an adjacency matrix of a common answer-interest network graph, INIs an identity matrix, HlRepresents the output of the l-1 th layer convolutional neural network, WlAnd a trainable weight matrix representing the l-th graph convolution neural network, wherein the output of the last layer of the graph convolution neural network is an airspace social characteristic vector.
6. The expert recommendation method in question-answering website according to claim 1, wherein the airspace social feature vectors are input into the full-link layer for classification to obtain classification feature vectors:
YU＝softmax(δ(W2Z+b2))
where δ represents a second nonlinear activation function, W1Trainable weight matrices representing fully connected layers of the first layer, XUIs a feature vector of the user, H is a spatial social feature vector, b1Is the offset vector of the first layer fully-connected layer, Z is the output of the first layer fully-connected layer, W2Trainable weight matrices representing fully connected layers of the second layer, b2Is the offset vector of the second layer fully-connected layer, YUIs a classification feature vector.
7. The expert recommendation method in question-answering website according to claim 1, further comprising the step of marking users meeting preset requirements as experts:
if the praise number of the user in a certain week is larger than the preset praise number threshold value, the user is marked as an expert in the week, and if the user does not answer the question in a certain week, the praise number in the week is continued.
8. An expert recommendation system in a question and answer website is characterized by comprising:
a time domain feature extraction module: the system comprises a data processing module, a data processing module and a data processing module, wherein the data processing module is used for acquiring question and answer characteristic vectors of all users and inputting the question and answer characteristic vectors into a pre-established gating circulation network model to obtain time domain semantic characteristic vectors;
the network graph building module: the system comprises a common answer-concern network graph and a common answer-concern network graph, wherein the common answer-concern network graph is constructed according to common answers and concern relations of users;
the spatial domain feature extraction module: the system comprises a time domain semantic feature vector input unit, a public answer-attention network graph learning unit and a public domain social feature vector input unit, wherein the time domain semantic feature vector is input into a pre-established graph convolution neural network to learn a topological structure of a common answer-attention network graph, and a spatial domain social feature vector is obtained;
an expert extraction module: the system comprises a spatial domain social characteristic vector input full-connection layer, a classification characteristic vector and an expert list, wherein the classification characteristic vector is obtained by inputting the spatial domain social characteristic vector into the full-connection layer for classification, and a user corresponding to an element with a median value of 1 in the classification characteristic vector is an expert;
a recommendation module: the system is used for obtaining the user characteristic vector and the specified problem characteristic vector of the expert, calculating the cosine similarity of the user characteristic vector and the specified problem characteristic vector of the expert, and taking the expert with the cosine similarity larger than the preset threshold value as a recommendation result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111469719.6A CN114329184A (en) | 2021-12-03 | 2021-12-03 | Expert recommendation method and system in question and answer website |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111469719.6A CN114329184A (en) | 2021-12-03 | 2021-12-03 | Expert recommendation method and system in question and answer website |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114329184A true CN114329184A (en) | 2022-04-12 |
Family
ID=81048280
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111469719.6A Pending CN114329184A (en) | 2021-12-03 | 2021-12-03 | Expert recommendation method and system in question and answer website |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114329184A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114896384A (en) * | 2022-06-14 | 2022-08-12 | 青岛科技大学 | Question answering expert cross-domain recommendation method and system for question answering community |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108021616A (en) * | 2017-11-06 | 2018-05-11 | 大连理工大学 | A kind of community's question and answer expert recommendation method based on Recognition with Recurrent Neural Network |
| CN110321421A (en) * | 2019-07-04 | 2019-10-11 | 南京邮电大学 | Expert recommendation method and computer storage medium for website Knowledge Community system |
| US20200167428A1 (en) * | 2018-11-26 | 2020-05-28 | International Business Machines Corporation | Utilizing external knowledge and memory networks in a question-answering system |
-
2021
- 2021-12-03 CN CN202111469719.6A patent/CN114329184A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108021616A (en) * | 2017-11-06 | 2018-05-11 | 大连理工大学 | A kind of community's question and answer expert recommendation method based on Recognition with Recurrent Neural Network |
| US20200167428A1 (en) * | 2018-11-26 | 2020-05-28 | International Business Machines Corporation | Utilizing external knowledge and memory networks in a question-answering system |
| CN110321421A (en) * | 2019-07-04 | 2019-10-11 | 南京邮电大学 | Expert recommendation method and computer storage medium for website Knowledge Community system |
Non-Patent Citations (2)
| Title |
|---|
| JINFENG XIAO 等: "Open-Domain Question Answering with Pre-Constructed Question Spaces", PROCEEDINGS OF NAACL-HLT 2021, 11 June 2021 (2021-06-11), pages 1 - 7 * |
| 刘臣 等: "基于图卷积神经网络的在线社区行为预测", 计算机技术与发展, vol. 31, no. 4, 30 April 2021 (2021-04-30), pages 1 - 7 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114896384A (en) * | 2022-06-14 | 2022-08-12 | 青岛科技大学 | Question answering expert cross-domain recommendation method and system for question answering community |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111538868B (en) | Knowledge tracking method and problem recommendation method | |
| CN106600052B (en) | User attribute and social network detection system based on space-time trajectory | |
| CN111190939A (en) | User portrait construction method and device | |
| CN111885399B (en) | Content distribution method, device, electronic equipment and storage medium | |
| CN111563192B (en) | Entity alignment method, device, electronic equipment and storage medium | |
| CN106997549A (en) | The method for pushing and system of a kind of advertising message | |
| CN104133817A (en) | Online community interaction method and device and online community platform | |
| CN110321421B (en) | Expert recommendation method for website knowledge community system and computer storage medium | |
| CN109903127A (en) | Group recommendation method and device, storage medium and server | |
| CN111506821B (en) | Recommendation model, recommendation method, recommendation device, recommendation equipment and recommendation storage medium | |
| CN114579584B (en) | Data table processing method and device, computer equipment and storage medium | |
| CN115456043A (en) | Classification model processing method, intent recognition method, device and computer equipment | |
| CN115588193A (en) | Visual question-answering method and device based on graph attention neural network and visual relation | |
| CN114116995B (en) | Session recommendation method, system and medium based on enhanced graph neural network | |
| CN114329184A (en) | Expert recommendation method and system in question and answer website | |
| CN112819154B (en) | Method and device for generating pre-training model applied to graph learning field | |
| CN106294406A (en) | A kind of method and apparatus accessing data for processing application | |
| CN113641791A (en) | Expert recommendation method, electronic device and storage medium | |
| CN113821676A (en) | Video retrieval method, device, equipment and storage medium | |
| CN116720009A (en) | Social robot detection method, device, equipment and storage medium | |
| CN111353001A (en) | Method and device for classifying users | |
| CN114692022A (en) | Position prediction method and system based on space-time behavior mode | |
| Alpaydin | Multiple neural networks and weighted voting | |
| CN114579744A (en) | Social network opinion leader mining method based on GraphSAGE | |
| CN113076450A (en) | Method and device for determining target recommendation list |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |