CN110443574B - Recommendation method for multi-project convolutional neural network review experts - Google Patents

Recommendation method for multi-project convolutional neural network review experts Download PDF

Info

Publication number
CN110443574B
CN110443574B CN201910677223.4A CN201910677223A CN110443574B CN 110443574 B CN110443574 B CN 110443574B CN 201910677223 A CN201910677223 A CN 201910677223A CN 110443574 B CN110443574 B CN 110443574B
Authority
CN
China
Prior art keywords
information data
expert
project
matrix
incidence relation
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.)
Active
Application number
CN201910677223.4A
Other languages
Chinese (zh)
Other versions
CN110443574A (en
Inventor
余正涛
王广祥
赖华
王剑
何孝胥
毛存礼
郭军军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Kunming University of Science and Technology
Original Assignee
Kunming University of Science and Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Kunming University of Science and Technology filed Critical Kunming University of Science and Technology
Priority to CN201910677223.4A priority Critical patent/CN110443574B/en
Publication of CN110443574A publication Critical patent/CN110443574A/en
Application granted granted Critical
Publication of CN110443574B publication Critical patent/CN110443574B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/10—Office automation; Time management
    • G06Q10/103—Workflow collaboration or project management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Engineering & Computer Science (AREA)
  • Marketing (AREA)
  • Biomedical Technology (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a recommendation method for multi-project convolutional neural network review experts, and belongs to the technical field of data processing. The invention analyzes various incidence relations between the project information data and between the expert project information data, expresses a plurality of project information data contents and a plurality of expert information data contents related to projects into vectors, constructs incidence relations between the project information data and the expert information data through a non-directional graph model, and respectively connects the incidence relations with the content vectors to form a matrix. And then, establishing an evaluation expert recommendation model fusing the project information data association relation and the expert information data association relation by adopting a convolutional neural network, wherein the model can learn the score relation between the project information data and the expert information data so as to be used for evaluating expert recommendation, and the recommendation method has a good effect.

Description

Recommendation method for multi-project convolutional neural network review experts
Technical Field
The invention relates to a recommendation method for multi-project convolutional neural network review experts, and belongs to the technical field of data processing.
Background
In recent years, the activities of declaration, establishment and the like of scientific and technological projects in China are increasing day by day, the information system is continuously realized by the management of scientific and technological projects and talents at all levels in China, and experts play a key role in review work all the time. In order to ensure the objectivity, fairness and fairness of the project review work, the selection work of the review experts is very important. The expert recommendation is to find an expert matched with a task to be recommended by utilizing technologies such as data mining, machine learning and the like based on rich expert database information. From the task to be recommended, expert recommendations can be seen as a special form of personalized recommendations based on content filtering. In the field of electronic commerce, recommendation algorithms and technologies are researched more, and a sufficient theoretical basis is provided for expert recommendation. Currently, in the research of recommendation methods, the following methods are mainly proposed, 1. Methods based on content; 2. collaborative filtering based methods; 3. a method based on deep learning.
How to fairly and automatically select the qualified review experts in activities such as the application and establishment of a large number of scientific and technological projects is always the key of project review work, and if a plurality of project review works are given, the review experts capable of directly selecting the project review works have very important application prospect. In the expert recommendation problem aiming at multi-project evaluation, not only the closeness of matching between the evaluation project content and the experts, but also various social relationships existing among the experts can influence the determination of the final evaluation expert choice, and meanwhile, fewer factors are considered in the conventional method. Therefore, how to automatically recommend the review experts for the plurality of items becomes one of the difficulties and key technologies of the task through the artificial intelligence technology by utilizing the contents of the plurality of items to be reviewed, the contents of the plurality of experts and the association relationship among the experts.
Disclosure of Invention
The invention provides a multi-project convolutional neural network review expert recommendation method, which expresses a plurality of project data contents and a plurality of expert data contents related to projects into vectors, constructs incidence relations between the project information data and the expert information data through a directionless graph model, and respectively connects the incidence relations with the content vectors to form a matrix; and then, establishing an evaluation expert recommendation model fusing the project information data association relation and the expert information data association relation by adopting a convolutional neural network to be applied to evaluation expert recommendation.
The technical scheme of the invention is as follows: the recommendation method for the multi-project convolutional neural network review experts comprises the following specific steps:
step1, data collection and vectorization representation: collecting project and expert information data, and expressing the project content information and the expert content information by vectors;
in the Step1, project and expert structured content data are collected, project content information and expert content information are characterized into vector representations which can be identified and processed by a computer through one-hot coding, and the vector representations are mapped into dense hidden vector representations through an Embedding layer.
Step2, construction of a project information data relation and an expert information data relation: extracting incidence relations existing between the project information data and between the expert information data and constructing an incidence relation matrix of the incidence relations;
as a preferable scheme of the invention, step2 represents the correlation between the project information data and the expert information data in the recommendation process by using the undirected edge of the Markov network, and constructs the Markov network through the correlation to obtain the incidence relation matrix between the project information data and the expert information data.
Step3, fusing the incidence relation matrix: on the basis of the Step1 and the Step2, respectively connecting the project information data matrix and the expert information data matrix with the incidence relation matrix;
as a preferred scheme of the invention, step3 vertically connects the hidden vectors obtained in Step1 into a project information data matrix and an expert information data matrix respectively, and then horizontally connects the obtained project information data matrix and the expert information data matrix with the project information data incidence relation matrix and the expert information data incidence relation matrix obtained in Step2 respectively into a project matrix with project information data incidence relation fused and an expert matrix with expert information data incidence relation fused.
Step4, constructing a recommendation model: and on the basis of Step3, constructing a review expert recommendation model by using a matrix fusing the project information data incidence relation and the expert information data incidence relation based on the convolutional neural network.
As a preferred embodiment of the present invention, in Step4, a convolutional neural network is selected, the item information data matrix fused with the item information data association relationship and the expert information data matrix fused with the expert information data association relationship are used as inputs to perform convolution and pooling, and the results obtained through the convolution and pooling are connected and learned to output a final score result, thereby implementing multi-item multi-expert recommendation.
The invention has the beneficial effects that:
the invention analyzes various incidence relations between the project information data and between the expert project information data, expresses a plurality of project information data contents and a plurality of expert information data contents related to projects into vectors, constructs incidence relations between the project information data and the expert information data through a non-directional graph model, and respectively connects the incidence relations with the content vectors to form a matrix. And then, establishing an evaluation expert recommendation model fusing the project information data association relation and the expert information data association relation by adopting a convolutional neural network, wherein the model can learn the score relation between the project information data and the expert information data, and realize multi-expert recommendation of a plurality of projects. Experimental results show that the recommendation method achieves better effect, and the recommendation effect is improved to a certain extent compared with a method without considering the relation between the project information data and the expert information data.
Drawings
FIG. 1 is a diagram of the correlation between review experts based on Markov network;
FIG. 2 is a convolutional neural network-based project review expert recommendation model proposed by the present invention;
fig. 3 is a block diagram of the proposed convolutional neural network-based multi-project review expert recommendation.
Detailed Description
Example 1: as shown in fig. 1-3, the method for recommending experts by a multi-project convolutional neural network review comprises the following specific steps:
step1, data collection and vectorization representation: and collecting and obtaining project and expert structured content data, converting the project and expert structured content data into binary sparse vectors by utilizing one-hot coding, and mapping the sparse vectors into dense hidden vector representations through an Embedding layer.
Step2, construction of project information data relation and expert information data relation: extracting incidence relations existing between the project information data and between the expert information data and constructing incidence relation matrixes of the incidence relations;
and extracting the incidence relation between the project information data and the expert information data, wherein the incidence relation characteristic of the expert information data is shown in a table 1, and the incidence relation characteristic of the project information data is shown in a table 2.
And constructing an expert Markov network by using the incidence relation characteristics among the expert information data to obtain an incidence relation matrix among the expert information data. In the expert Markov network, the correlation between expert information data is calculated by fusing the association relation of experts. In order to fuse the expert information data incidence relations of the types, the expert information data incidence relations are defined as a characteristic function h m (e i ,e j ) And assigning a weight λ to each eigenfunction m And fusing the correlation characteristics by using a logarithmic linear model. Sim (e) for correlation between expert information data is assumed i ,e j ) Expressed, the correlation calculation formula is as follows:
Figure BDA0002143660370000031
wherein the content of the first and second substances,
Figure BDA0002143660370000032
it represents the mth characteristic function, M represents the total number of the associated characteristic, lambda m Represents the weight of the corresponding characteristic function>
Figure BDA0002143660370000033
Denotes all and nodes e i A set of connected edges. Characteristic function weight lambda m And estimating the data by adopting a maximum likelihood estimation method. When Sim (e) i ,e j ) Is large in valueAt a given threshold value beta, the expert information data e is considered i And expert information data e j It is relevant that there is a undirected edge between these two nodes in the expert Markov network, as shown in FIG. 1, the edge having a weight of Sim (e) i ,e j ). And similarly, constructing an incidence relation matrix among the project information data.
TABLE 1 expert information data Association relation features
Figure BDA0002143660370000041
TABLE 2 item information data Association relationship features
Figure BDA0002143660370000042
Step3, fusing the incidence relation matrix: on the basis of the Step1 and the Step2, respectively connecting the project information data matrix and the expert information data matrix with the incidence relation matrix;
as a preferred scheme of the present invention, step3 vertically connects the hidden vectors obtained in Step1 to form a project information data matrix and an expert information data matrix, and then horizontally connects the obtained project information data matrix and the expert information data matrix to the project information data incidence relation matrix and the expert information data incidence relation matrix obtained in Step2 to form a project matrix fused with the project information data incidence relation and an expert matrix fused with the expert information data incidence relation.
Specifically, the hidden vectors obtained after Embedding are respectively and vertically connected into an n × k item matrix and an m × l expert matrix, where n is the number of items, k is the dimension of the hidden vectors of the items, m is the number of experts, and l is the dimension of the hidden vectors of the experts.
And horizontally connecting the obtained n multiplied by k item matrix and m multiplied by l expert matrix with the n multiplied by n relation matrix of the n items and the m multiplied by m relation matrix of the m items into an item matrix fused with the item information data incidence relation and an expert matrix fused with the expert information data incidence relation.
Step4, establishing a recommendation model: and on the basis of Step3, constructing a review expert recommendation model by using a matrix fusing the project information data incidence relation and the expert information data incidence relation based on the convolutional neural network.
As a preferred scheme of the present invention, a convolutional neural network fusion project information data relationship and expert information data association relationship are adopted to construct a review expert recommendation model in combination with a computer technology, as shown in fig. 2, the implementation of the convolutional neural network expert recommendation model specifically comprises the following steps:
and Step4.1, performing convolution on the project joint matrix fused with the project information data incidence relation and the expert joint matrix fused with the expert information data incidence relation by utilizing a convolution neural network so as to extract characteristics. Let x i ∈R n An n-dimensional vector representing the ith item or expert in the item-joint matrix or expert-joint matrix, the matrix containing m experts can be represented as:
Figure BDA0002143660370000051
wherein the content of the first and second substances,
Figure BDA0002143660370000052
indicating a join operation, a convolution operation from x using a sliding window of length h 1:m And intercepting the vector as the input of the activation function of the convolution layer, and calculating the activation function to obtain the output which is the feature to be extracted. As shown in the following formula:
c i ＝f(W·x i:i+h-1 +b)
where the function f represents an activation function, the present invention selects the ReLU function as the activation function. W and b represent a weight matrix and a bias unit, are parameters of the neural network and can be obtained by training the neural network. c. C i Representing the input vector x according to an activation function f i:i+h-1 The calculated features.
The above formula describes the sliding window of length h at position x i:i+h-1 The above-described operations, when the sliding window slides from the beginning to the end of a sentence, result in a set of inputs, denoted as (x) 1:h ,x 2:1+h ,...,x m-h+1:m ) Further, a set of outputs, denoted as (c), may be derived from the inputs 1 ,c 2 ,...,c m-h+1 ) This set of outputs is called a feature map.
And Step4.2, processing the feature mapping of the last step through a max-posing layer to obtain a final feature vector. In particular, the maximum values in the feature maps are extracted, the idea being to consider that the maximum value in each feature map represents the most important feature in the map. Furthermore, one of the main advantages of the max-posing layer is that a fixed-length feature vector can be obtained by means of the operation of this layer, without having to care about the length of the original input vector.
And Step4.3, connecting and learning the results obtained by the convolution and pooling of the project information data matrix and the expert information data matrix and outputting the final score result. Specifically, in the MLP layer, a Relu function is used as an activation function, and softmax processing is carried out on the result to obtain the probability. If the softmax layer category number, namely the number of neurons, is k, then:
Figure BDA0002143660370000053
wherein z represents an input vector of the softmax layer, W represents a parameter of the softmax layer network, and y i And the output value of the ith neuron of the output layer is represented, and the final output result is the probability value of k types of vectors represented by k types. The output layer of the model determines the number of neurons from the number of classes of the classification task. In order to obtain the probability values of the score sizes of the n experts for the same item, n softmax are used for outputting n corresponding score results at the last layer of the MLP layer.
In order to explain the performance of the invention, a model which does not use a project information data incidence relation matrix and an expert information data incidence relation matrix is selected for comparison, a method that an input layer does not use the project information data incidence relation matrix and the expert information data incidence relation matrix as auxiliary input is marked as M1, and a method that the input layer uses the project information data incidence relation and the expert information data incidence relation matrix as auxiliary input is marked as M2;
and connecting and learning the results obtained by convolution and pooling, outputting the final average NDCG value of different methods, and comparing and analyzing the expert recommendation result through the NDCG value to verify the optimal method recommended by the expert.
Table 3 is the average NDCG values for different methods (where K is the dimension size after Embedding for each type of attribute set in the expert or project information data content representation).
Table 3: average NDCG values for different methods
K M1 M2
8 0.901 0.905
16 0.913 0.919
32 0.918 0.920
As can be seen from the above data, the effect obtained by using the expert information data association relationship and the project information data association relationship as the auxiliary input is better than that obtained by using the model without using the expert information data association relationship and the project information data association relationship as the auxiliary input. Experimental results prove that the method has a good effect on the NDCG index sorted by the recommendation system.
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.

Claims (2)

1. The recommendation method for the multi-project convolutional neural network review experts is characterized by comprising the following steps: the method comprises the following specific steps:
step1, data collection and vectorization representation: collecting project and expert information data, and expressing project content information and expert content information by vectors;
step2, construction of a project information data relation and an expert information data relation: extracting incidence relations existing between the project information data and between the expert information data and constructing incidence relation matrixes of the incidence relations;
step3, fusing the incidence relation matrix: on the basis of the Step1 and the Step2, respectively connecting the project information data matrix and the expert information data matrix with the incidence relation matrix;
step4, constructing a recommendation model: on the basis of Step3, constructing a review expert recommendation model by using a matrix fusing the incidence relation of the project information data and the incidence relation of the expert information data based on the convolutional neural network;
in Step1, collecting and obtaining project and expert structured content data, representing project content information and expert content information into vector representations which can be identified and processed by a computer through one-hot coding, and mapping the vector representations into dense hidden vector representations through an Embedding layer;
step3, respectively and vertically connecting the hidden vectors obtained in Step1 into a project information data matrix and an expert information data matrix, and then horizontally connecting the obtained project information data matrix and the expert information data matrix with the project information data incidence relation matrix and the expert information data incidence relation matrix obtained in Step2 into a project matrix fused with the project information data incidence relation and an expert matrix fused with the expert information data incidence relation;
and selecting a convolutional neural network in the Step4, performing convolution and pooling by using the project information data matrix fused with the project information data association relation and the expert information data matrix fused with the expert information data association relation as input, connecting and learning the results obtained through the convolution and pooling, and outputting a final score result, thereby realizing multi-project multi-expert recommendation.
2. The multi-project convolutional neural network review expert recommendation method of claim 1, wherein: and Step2, representing the correlation between the project information data and the expert information data in the recommendation process by using the undirected edges of the Markov network, and constructing the Markov network through the correlation to obtain an incidence relation matrix between the project information data and the expert information data.
CN201910677223.4A 2019-07-25 2019-07-25 Recommendation method for multi-project convolutional neural network review experts Active CN110443574B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910677223.4A CN110443574B (en) 2019-07-25 2019-07-25 Recommendation method for multi-project convolutional neural network review experts

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910677223.4A CN110443574B (en) 2019-07-25 2019-07-25 Recommendation method for multi-project convolutional neural network review experts

Publications (2)

Publication Number Publication Date
CN110443574A CN110443574A (en) 2019-11-12
CN110443574B true CN110443574B (en) 2023-04-07

Family

ID=68431577

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910677223.4A Active CN110443574B (en) 2019-07-25 2019-07-25 Recommendation method for multi-project convolutional neural network review experts

Country Status (1)

Country Link
CN (1) CN110443574B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11200591B2 (en) * 2018-05-30 2021-12-14 Paypal, Inc. Electronic content based on neural networks
CN111782797A (en) * 2020-07-13 2020-10-16 贵州省科技信息中心 Automatic matching method for scientific and technological project review experts and storage medium
CN113254642A (en) * 2021-05-28 2021-08-13 华斌 E-government affair project evaluation expert group recommendation method based on multi-dimensional feature balance

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103136267A (en) * 2011-12-01 2013-06-05 腾讯科技（深圳）有限公司 Expert mining method and device based on community
CN103823896A (en) * 2014-03-13 2014-05-28 蚌埠医学院 Subject characteristic value algorithm and subject characteristic value algorithm-based project evaluation expert recommendation algorithm
CN109446413A (en) * 2018-09-25 2019-03-08 上海交通大学 Serializing recommended method based on item associations relationship

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103631859B (en) * 2013-10-24 2017-01-11 杭州电子科技大学 Intelligent review expert recommending method for science and technology projects
CN108804689B (en) * 2018-06-14 2020-10-16 合肥工业大学 Question-answering platform-oriented label recommendation method integrating user hidden connection relation
CN109241424B (en) * 2018-08-29 2019-08-27 陕西师范大学 A kind of recommended method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103136267A (en) * 2011-12-01 2013-06-05 腾讯科技（深圳）有限公司 Expert mining method and device based on community
CN103823896A (en) * 2014-03-13 2014-05-28 蚌埠医学院 Subject characteristic value algorithm and subject characteristic value algorithm-based project evaluation expert recommendation algorithm
CN109446413A (en) * 2018-09-25 2019-03-08 上海交通大学 Serializing recommended method based on item associations relationship

Also Published As

Publication number Publication date
CN110443574A (en) 2019-11-12

Similar Documents

Publication Publication Date Title
CN111523047B (en) Multi-relation collaborative filtering algorithm based on graph neural network
CN108920720B (en) Large-scale image retrieval method based on depth hash and GPU acceleration
CN112529168B (en) GCN-based attribute multilayer network representation learning method
CN111061856B (en) Knowledge perception-based news recommendation method
CN112232925A (en) Method for carrying out personalized recommendation on commodities by fusing knowledge maps
CN109389151B (en) Knowledge graph processing method and device based on semi-supervised embedded representation model
CN112115377B (en) Graph neural network link prediction recommendation method based on social relationship
CN110443574B (en) Recommendation method for multi-project convolutional neural network review experts
CN112364976B (en) User preference prediction method based on session recommendation system
Huang et al. Forecasting container throughput of Qingdao port with a hybrid model
CN110442618B (en) Convolutional neural network review expert recommendation method fusing expert information association relation
CN105512273A (en) Image retrieval method based on variable-length depth hash learning
CN112508085A (en) Social network link prediction method based on perceptual neural network
CN110837602A (en) User recommendation method based on representation learning and multi-mode convolutional neural network
CN112559764A (en) Content recommendation method based on domain knowledge graph
CN113705099A (en) Social platform rumor detection model construction method and detection method based on contrast learning
CN112905906B (en) Recommendation method and system fusing local collaboration and feature intersection
CN113837492B (en) Method, apparatus, storage medium, and program product for predicting article supply amount
CN111340637A (en) Medical insurance intelligent auditing system based on machine learning feedback rule enhancement
CN115861664A (en) Feature matching method and system based on local feature fusion and self-attention mechanism
CN113283243B (en) Entity and relationship combined extraction method
CN114297498A (en) Opinion leader identification method and device based on key propagation structure perception
Bandyopadhyay et al. Sac2vec: Information network representation with structure and content
CN114494753A (en) Clustering method, clustering device, electronic equipment and computer-readable storage medium
CN112016004A (en) Multi-granularity information fusion-based job crime screening system and method

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
GR01 Patent grant
GR01 Patent grant