CN112508256B - User demand active prediction method and system based on crowdsourcing - Google Patents
User demand active prediction method and system based on crowdsourcing Download PDFInfo
- Publication number
- CN112508256B CN112508256B CN202011387991.5A CN202011387991A CN112508256B CN 112508256 B CN112508256 B CN 112508256B CN 202011387991 A CN202011387991 A CN 202011387991A CN 112508256 B CN112508256 B CN 112508256B
- Authority
- CN
- China
- Prior art keywords
- user
- demand
- crowdsourcing
- information
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 230000006399 behavior Effects 0.000 claims abstract description 23
- 239000013604 expression vector Substances 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims abstract description 8
- 239000013598 vector Substances 0.000 claims description 126
- 239000011159 matrix material Substances 0.000 claims description 72
- 230000006870 function Effects 0.000 claims description 31
- 230000014509 gene expression Effects 0.000 claims description 14
- 230000004913 activation Effects 0.000 claims description 12
- 230000003993 interaction Effects 0.000 claims description 11
- 230000004931 aggregating effect Effects 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 9
- 230000002452 interceptive effect Effects 0.000 claims description 7
- 230000007246 mechanism Effects 0.000 claims description 7
- 239000000126 substance Substances 0.000 claims description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000011478 gradient descent method Methods 0.000 claims description 6
- 238000012163 sequencing technique Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 4
- 238000004519 manufacturing process Methods 0.000 abstract description 3
- 238000005457 optimization Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06315—Needs-based resource requirements planning or analysis
-
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Entrepreneurship & Innovation (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Game Theory and Decision Science (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Educational Administration (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a user demand active prediction method and a user demand active prediction system based on crowdsourcing, which comprise the following steps: s1: determining annotators participating in the crowdsourcing task, wherein the annotators receive the crowdsourcing task and complete the task; s2: constructing a heterogeneous information network according to the user preference information; s3: generating a user required data space; s4: learning the expression vectors of the user and the demand object respectively through a graph convolution neural network; s5: and (4) demand prediction. According to the invention, the user directly participates in information production and knowledge sharing through a crowdsourcing technology, the preference information fed back by a crowdsourcing annotator can better reflect the real requirement of the user, and the accuracy of the result can be improved by combining the information to predict the requirement; the attribute characteristics of the users are enriched by the user preference information acquired in the crowdsourcing mode, and attribute completion is performed on new registered users lacking historical behavior data, so that each user can be more accurately represented, and the recommendation result is more personalized.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a user demand active prediction method and system based on crowdsourcing.
Background
With the development of the internet and big data technology, the problem of information overload is increasingly serious, and the recommendation system can provide interested commodities or services for the user according to the demand information of the user to help the user to perform effective information processing. Therefore, whether the user requirements can be accurately, comprehensively and actively predicted becomes a key for improving the recommendation performance of the service provider so as to realize the maximization of the commercial profit.
(1) Heterogeneous information network recommendation
The existing recommendation technology mainly analyzes user requirements based on network behavior data of users, and the user network behavior data under the background of a big data era often has multi-source heterogeneity. Heterogeneous information networks can integrate different types of objects and complex interaction relationships between the objects, and have been widely used in the field of recommendation as an effective information fusion method. The application number CN106503028A relates to a recommendation method, which comprises the following steps: modeling the objects in the recommended data set and the relationship between the objects as a heterogeneous information network; acquiring a meta path connecting two objects in the heterogeneous information network; calculating similarity data between the objects according to a meta path connecting the two objects; constructing an objective function according to similarity data between objects, and training the recommended data set through the objective function to obtain a prediction score of the user on the object; and recommending the item to the user according to the prediction score of the user on the item. According to the method, a heterogeneous information network is utilized to model the recommendation data, the problem of data sparseness is effectively relieved, and the recommendation effect is improved.
(2) Crowdsourcing recommendations
Crowdsourcing is a distributed problem solving and production mode which adopts a certain mechanism to enable groups to participate in a certain thing together to achieve a certain target. Crowdsourcing solves the problem difficult to understand through the intelligence of the group, and spreads the problem to the worker group in a public bidding mode. Combining crowdsourcing task information has become a hot issue for research in the field of crowdsourcing recommendations.
A method for crowd-sourced task recommendation, as disclosed in application No. 202010464312.3, comprising the steps of: according to crowdsourcing worker data and historical tasks on a crowdsourcing platform, user portrait updating and user portrait grade updating are carried out on crowdsourcing workers; screening the crowdsourcing workers according to the requirements of the tasks to be processed, and obtaining a crowdsourcing worker list; determining the completion time and price of the task to be processed according to the requirements of the task to be processed and the crowdsourcing worker list; determining a recommendation probability list of the crowdsourcing workers through a task recommendation model according to the completion time and the price; recommending the tasks to be processed to crowdsourcing workers in the crowdsourcing worker list according to the tasks to be processed and the recommendation probability list. According to the method and the device, the crowdsourcing workers are subjected to user portrait, skills of the crowdsourcing workers are graded according to attributes in the user portrait, and the recommendation probability list is generated, so that tasks are automatically pushed to the crowdsourcing workers. The method focuses on task recommendation on a crowdsourcing platform. The crowdsourcing mode can feed back information which can reflect true requirements of the user to a task requester in a mode of direct participation of the user, such as a service provider, more accurate and complete data support can be provided for user requirement prediction of a service platform by combining data collected by the crowdsourcing mode with recommended data, and particularly the data missing problem of a newly registered service platform user can be relieved. Therefore, designing a heterogeneous information network modeling method fusing crowdsourcing acquisition data and recommendation data and a corresponding demand prediction method is a practical demand.
Disclosure of Invention
The invention aims to provide a high-matching-degree demand active prediction method for a new user lacking historical data.
The invention solves the technical problems through the following technical means:
a crowd-sourced user demand active prediction method comprises the following steps:
s1: determining annotators participating in the crowdsourcing task, designing the crowdsourcing task and issuing the crowdsourcing task to a crowdsourcing task platform, wherein the annotators receive the crowdsourcing task and complete the task;
s2: constructing a heterogeneous information network according to the user social relationship, the historical behavior data and the user preference information acquired in the crowdsourcing mode;
s3: uniformly representing different types of entities in a heterogeneous information network to generate a user demand data space;
s4: extracting the interactive semantics of the user and the demand object by using a meta path in a heterogeneous information network, and respectively learning the expression vectors of the user and the demand object through a graph convolution neural network;
s5: and aggregating the neighbor information of the target user according to the social relation of the users in the heterogeneous information network, obtaining the expression vector of the target user from the data space, and predicting the demand.
Further, the step S1 comprises
S11: acquiring a user set of a service provider as a target user, acquiring user social relationship data from a social network and a service provider platform to obtain a social neighbor user set of the target user, and taking all users as annotators for receiving crowdsourcing tasks;
s12: and designing a user preference survey questionnaire from the aspects of demographic information, social requirements for reflecting common interest in social relations and enjoyment requirements for reflecting individual preferences, wherein the content of the questionnaire comprises selection questions expressed in characters and displayed in a graphical mode, a annotator is allowed to submit auxiliary information independently, the questionnaire is published on a crowdsourcing platform in a crowdsourcing task mode, and tasks are published to the annotator obtained in S11.
Further, the step S2 includes:
s21: taking multi-modal data acquired based on a crowdsourcing mode in the S1 as an attribute set of each user, and taking the user and a demand object as nodes;
s22: establishing a connection edge between a user and a demand object according to the following relation:
the relationship R1: direct relationships such as friends and concern exist among users U, and L are used respectively -1 Representing relationships between users U, i.e.
And &>
Relationship R2: some users have historical behavior information, such as the user bought a certain article, used a certain service, etc., respectively using B and B -1 Representing user U and requirement object O k In relation to each other, i.e.
And &>
Wherein k represents a kth class requirement object;
s23: and establishing a multi-mode heterogeneous information network according to the attribute set, the nodes and the relationship among the nodes.
Further, the step S3 includes:
s31: user information, text attribute information and image attribute information of a demand object collected in a crowdsourcing mode are uniformly expressed:
obtaining vector representation of text type information by adopting word2vec method
Wherein e is u Representing a user text attribute vector representation, e o Representing the text attribute vector representation of the demand object, wherein N is the quantity of the demand object categories;
the picture type information is represented by vector obtained by adopting convolutional neural network
Wherein, g u Representing user Picture Attribute vector representation, g o A picture attribute vector representation representing a demand object;
s32: to the warpFusing the uniformly expressed multi-modal attribute information: the user attribute vector e obtained in the step S31 u And g u Performing outer product operation to realize feature intersection, flattening the obtained matrix according to rows, inputting the flattened matrix into a multilayer perceptron to obtain an initial vector representation Z of a user node, and expressing a vector of a demand object attribute
And &>
Repeating the operation to obtain the initial vector representation O of the demand object node k And vector representation of all users and requirement objects forms a user requirement data space.
Further, the step S4 includes:
s41: establishing a plurality of user-demand object co-occurrence matrixes T according to historical behavior information of users k : user-item co-occurrence matrix
Subscriber-service co-occurrence matrix ≥>
Wherein, | I | is the quantity of articles, | S | is the quantity of services, if the user has bought a certain article or the user has used a certain service, put 1 in the corresponding position of the corresponding co-occurrence matrix;
s42: in the k-th demand active prediction scene of the user, extracting the UO from the heterogeneous information network constructed in the step S2 k U-element path, meaning that two users use the semantic information of the kth class demand object together, co-occurrence matrix T k To which it is transferred
Multiply, i.e. ->
Deriving an inter-user relationship matrix under that semantic->
Extracting O from the heterogeneous information network constructed in step S2 k UO k Meta-path representing semantic information used by the same user for two kth class demand objects by &>
Obtain a relationship matrix ^ between the kth class of demand objects under the semantic>
S43: for the relationship matrix between users obtained in S42
And a relationship matrix between demand objects>
The standardization treatment is carried out according to the following formulas respectively,
wherein, the first and the second end of the pipe are connected with each other,
and &>
Are all diagonal matrices, are asserted>
And &>
Are respectively based on>
And &>
A degree matrix of (c);
s44: using a graph-convolution neural network, a user vector representation is learned according to the following formula,
the vector representation of the kth class demand object is learned in accordance with the following formula,
wherein the content of the first and second substances,
vector representations of the ith layer user and kth class requirement object respectively are shown, when l is 0,
is Z, <' > based on>
Is O k P and W are weight parameters, wherein, indicates element-by-element multiplication operation, sigma is an activation function, and phi indicates that a vector is converted into a diagonal matrix;
s45: and repeating the operation in the S44, and alternately updating the vector representations of the user and the kth class of demand object respectively until the final layer of convolution is finished to obtain the vector representations of all the user and the kth class of demand object.
Further, the step S5 includes:
s51: s4, obtaining a vector representation of a target user i as
Neighbor user j e of user iN (i), aggregating neighbor user information by using an attention mechanism to obtain final target user vector representation; the weight coefficient of the neighbor to the target user is calculated,
the vector representation of the target user is updated,
wherein, alpha and W are weight parameters, sigma is an activation function, and | l is splicing operation;
s52: for each target user, calculating its relevance prediction score to each kth class demand object
S53: the loss function is a binary cross entropy function:
wherein, Y and Y - Positive and negative examples in the data set, Y represents the demand object set used by the user, Y - Sampled from the demand objects in the data set that are not used by the user,
indicates whether the user has an interaction, presence interaction @, with the demand object>
Is 1, otherwise is 0; using stochastic gradient descent method to correct the loss functionPerforming optimization solution on the numbers, sequencing the kth class demand objects from high to low according to the prediction score obtained by calculation in the step S52, and selecting the first n demand objects as a kth class demand list of the user;
s54: by repeating the operations of S42-S53, a list of all the categories of demand objects for each user can be obtained, thereby realizing active prediction of user demands.
The invention also provides a user demand active prediction system based on crowdsourcing, which comprises
A crowdsourcing task issuing module: determining a annotator participating in the crowdsourcing task, designing the crowdsourcing task and issuing the crowdsourcing task to a crowdsourcing task platform, and receiving the crowdsourcing task and completing the task by the annotator;
heterogeneous information network construction module: constructing a heterogeneous information network according to the user social relationship, the historical behavior data and the user preference information acquired in the crowdsourcing mode;
the user demand data space generation module: uniformly representing different types of entities in a heterogeneous information network to generate a user demand data space;
the user and demand object representation vector learning module: extracting the interactive semantics of the user and the demand object by using a meta path in the heterogeneous information network, and respectively learning the expression vectors of the user and the demand object;
a demand forecasting module: and aggregating the neighbor information of the target user according to the social relation of the users in the heterogeneous information network to obtain the expression vector of the target user and perform demand prediction.
Further, the specific execution process of the step crowdsourcing task issuing module is as follows:
s11: acquiring a user set of a service provider as a target user, acquiring user social relationship data from a social network and a service provider platform to obtain a social neighbor user set of the target user, and taking all the users as annotators for receiving crowdsourcing tasks;
s12: and designing a user preference survey questionnaire from the aspects of demographic information, social requirements for reflecting common interest in social relations and enjoyment requirements for reflecting individual preferences, wherein the content of the questionnaire comprises selection questions expressed in characters and displayed in a graphical mode, a annotator is allowed to submit auxiliary information independently, the questionnaire is published on a crowdsourcing platform in a crowdsourcing task mode, and tasks are published to the annotator obtained in S11.
Further, the heterogeneous information network construction module performs the following steps:
s21: taking multimodal data collected based on a crowdsourcing mode as an attribute set of each user, and taking the user and a demand object as nodes;
s22: establishing a connection edge between a user and a demand object according to the following relation:
the relationship R1: direct relationships such as friends and concerns exist between users U, and L are used respectively -1 Representing relationships between users U, i.e.
And &>
The relationship R2: some users have historical behavior information, such as the user bought a certain article, used a certain service, etc., respectively using B and B -1 Representing user U and requirement object O k In relation to each other, i.e.
And &>
Wherein k represents a kth class requirement object;
s23: and establishing a multi-mode heterogeneous information network according to the attribute set, the nodes and the relationship among the nodes.
Further, the specific execution process of the user requirement data space generation module includes:
s31: user information, text attribute information and image attribute information of a demand object collected in a crowdsourcing mode are uniformly expressed:
obtaining vector representation of text type information by adopting word2vec method
Wherein e is u Representing a user text attribute vector representation, e o Representing the text attribute vector representation of the demand object, wherein N is the quantity of the demand object categories;
the picture type information is represented by vector obtained by adopting convolutional neural network
Wherein, g u Representing user picture attribute vector representation, g o A picture attribute vector representation representing a demand object;
s32: fusing the multi-mode attribute information after uniform expression: the user attribute vector e obtained in S31 is used u And g u Performing outer product operation to realize feature crossing, flattening the obtained matrix according to rows, and inputting the flattened matrix into a multilayer perceptron to obtain a user node initial vector representation Z and a demand object attribute representation vector
And &>
Repeating the operation to obtain the initial vector representation O of the demand object node k And vector representation of all users and requirement objects forms a user requirement data space.
Further, the specific implementation process of the expression vector learning module for the user and the demand object includes:
s41: establishing a plurality of user-demand object co-occurrence matrixes T according to historical behavior information of users k : user-item co-occurrence matrix
Subscriber-service co-occurrence matrix ≥>
Wherein, | I | is the quantity of articles, | S | is the quantity of services, if the user has bought a certain article or the user has used a certain service, put 1 in the corresponding position of the corresponding co-occurrence matrix;
s42: in the k-th demand active prediction scene of the user, extracting the UO from the constructed heterogeneous information network k U-element path, meaning that two users commonly use semantic information of kth class demand object, co-occurrence matrix T k To which it is transferred
Multiply, i.e. ->
Deriving an inter-user relationship matrix under that semantic->
Extracting O from a constructed heterogeneous information network k UO k Meta-path representing semantic information used by the same user for two kth class demand objects by &>
Obtain a relationship matrix ^ between the kth class of demand objects under the semantic>
/>
S43: for the relationship matrix between users obtained in S42
And a relation matrix between demand objects>
The standardization treatment is carried out according to the following formulas respectively,
wherein the content of the first and second substances,
and &>
Are all diagonal matrices, are asserted>
And &>
Are respectively based on>
And &>
A degree matrix of (c);
s44: using a graph convolution neural network, a user vector representation is learned in accordance with the following formula,
the vector representation of the kth class demand object is learned as follows,
wherein the content of the first and second substances,
vector representations of the ith layer user and kth class requirement object respectively are shown, when l is 0,
is Z->
Is O k P and W are weight parameters, wherein, indicates element-by-element multiplication operation, sigma is an activation function, and phi indicates that a vector is converted into a diagonal matrix;
s45: and repeating the operation in the S44, and alternately updating the vector representations of the user and the kth class of demand object respectively until the final layer of convolution is finished to obtain the vector representations of all the user and the kth class of demand object.
Further, the specific implementation process of the step demand prediction module includes:
s51: the expression vector learning module of the user and the demand object obtains the vector expression of the target user i as
Neighbor users j of the user i belong to N (i), and the neighbor user information is aggregated by using an attention mechanism to obtain final target user vector representation; the weight coefficient of the neighbor to the target user is calculated,
the vector representation of the target user is updated,
wherein, alpha and W are weight parameters, sigma is an activation function, and | l is splicing operation;
s52: for each target user, calculating the relevance prediction score of the target user and each k-th class demand object
S53: the loss function is a binary cross entropy function:
wherein Y and Y - Positive and negative examples in the data set, Y represents the demand object set used by the user, Y - Sampled from the demand objects in the data set that are not used by the user,
indicates whether a user has interacted with the demand object, there is an interaction @>
Is 1, otherwise is 0; optimizing and solving the loss function by using a random gradient descent method, sequencing kth demand objects from high to low according to the prediction score calculated in the step S52, and selecting the first n demand objects as a kth demand list of the user; />
S54: by repeating the operations of S42-S53, a list of all the categories of demand objects for each user can be obtained, thereby realizing the active prediction of the demand of the user.
The invention has the advantages that:
according to the invention, the user directly participates in information production and knowledge sharing through a crowdsourcing technology, the preference information fed back by a crowdsourcing annotator can better reflect the real requirement of the user, and the accuracy of the result can be improved by combining the information to predict the requirement; the attribute characteristics of the users are enriched by the user preference information acquired in the crowdsourcing mode, and attribute completion is performed on new registered users lacking historical behavior data, so that each user can be more accurately represented, and the recommendation result is more personalized.
The social relationship, the user historical behavior data and the user preference information collected in the crowdsourcing mode generally have the characteristics of multiple sources, multiple types and multiple relationships, the heterogeneous information network can effectively model multiple types of entities and complex relationships among the entities, and implicit interactive relationships among the entities of different types are extracted through meta-paths, so that the information is fully utilized, the individual characteristics of users and demand objects can be more comprehensively described, and the potential demands of the users are mined.
Drawings
Fig. 1 is a flow chart of a user demand active prediction method based on crowdsourcing in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, the present embodiment provides a method for actively predicting user demand based on crowdsourcing, which includes the following steps:
s1: determining annotators participating in the crowdsourcing task, designing the crowdsourcing task and issuing the crowdsourcing task to a crowdsourcing task platform; specifically comprises
S11: acquiring a user set of a service provider as a target user, acquiring user social relationship data from Twitter, a service provider platform and the like to obtain a social neighbor user set of the target user, and taking all the users as annotators for receiving crowdsourcing tasks;
s12: a user preference survey questionnaire is designed from the aspects of demographic information, social requirements reflecting common interest in social relations, enjoyment requirements reflecting personal preferences and the like, wherein the content of the questionnaire comprises word expressions and selection questions displayed graphically (for example, the interested content is selected from the following options), and a annotator is allowed to independently submit auxiliary information such as texts, videos and pictures. Issuing the questionnaire to a crowdsourcing platform in a crowdsourcing task form, and issuing a task to the annotator acquired in S11;
s2: constructing a heterogeneous information network according to the user social relationship, the historical behavior data and the user preference information acquired in the crowdsourcing mode; specifically comprises
S21: and taking the multi-modal data collected based on the crowdsourcing mode in the S1 as a property set of each user, such as age, gender, favorite movie posters and the like, and the properties of a demand object, such as a manufacturer, a marketing date and the like. The user and demand objects (including goods, services, etc.) are taken as nodes.
S22: establishing a connection edge between a user and a demand object according to the following relation:
the relationship R1: direct relationships such as friends and concern exist among users, and L are used respectively -1 Representing relationships between users (U), i.e.
And &>
Relationship R2: some users have historical behavior information, such as the user bought a certain article, used a certain service, etc., and use B and B respectively -1 Representing user (U) and requirement object (O) k ) In relation to each other, i.e.
And &>
Wherein k represents a kth-class demand object, such as a commodity or a service;
s23: and establishing a multi-mode heterogeneous information network according to the attribute set, the nodes and the relationship among the nodes.
S3: uniformly representing different types of entities in a heterogeneous information network to generate a user demand data space;
s31: user information and attribute information of demand objects collected based on crowdsourcing mode generally have different expression forms including types of texts, pictures and the like, and the user information and the attribute information of the demand objects need to be acquired according to different modalities without adopting the modesThe same expression learning method is used for unified expression. Obtaining vector representation of text type information by adopting word2vec method
e u Representing a user text attribute vector representation, e o And representing the text attribute vector representation of the demand object, wherein N is the number of the demand object categories. The picture type information is expressed by a vector obtained by a convolutional neural network>
g u Representing user picture attribute vector representation, g o A picture attribute vector representation representing a demand object;
s32: in order to learn the embedded representation of the nodes, the multimodality attribute information after being uniformly expressed needs to be fused. The user attribute vector e obtained in S31 is used u And g u Performing outer product operation to realize feature intersection, flattening the obtained matrix according to rows, inputting the flattened matrix into a multilayer perceptron to obtain an initial vector representation Z of a user node, and expressing a vector of a demand object attribute
And &>
Repeating the operation to obtain the initial vector representation O of the demand object node k The vector representations of all users and demand objects form a user demand data space.
S4: extracting the interactive semantics of the user and the demand object, and respectively learning the expression vectors of the user and the demand object; specifically comprises
S41: establishing a plurality of user-demand object co-occurrence matrixes T according to historical behavior information of users k E.g. user-item co-occurrence matrix
Subscriber-service co-occurrence matrix ≥>
Wherein, | I | is the quantity of articles, | S | is the quantity of services, if the user has bought a certain article or the user has used a certain service, put 1 in the corresponding position of the corresponding co-occurrence matrix;
s42: in the k-th demand active prediction scene of the user, extracting UO from the heterogeneous information network constructed by S2 k U-element path, meaning that two users commonly use semantic information of kth class demand object, co-occurrence matrix T k And the transpose thereof
Multiply, i.e. ->
Get the inter-user relationship matrix->
Extracting O from S2-constructed heterogeneous information network k UO k Meta-path representing semantic information used by the same user for two kth class demand objects by &>
Deriving a relationship matrix between class k demand objects under the semantics>
S43: for the relationship matrix between users obtained in S42
And a relation matrix between demand objects>
The standardization treatment is carried out according to the following formulas respectively,
wherein the content of the first and second substances,
and &>
Are all diagonal matrices, in combination>
And &>
Are respectively based on>
And &>
A degree matrix of (c);
s44: using a graph convolution neural network, a user vector representation is learned in accordance with the following formula,
the vector representation of the kth class demand object is learned in accordance with the following formula,
wherein, the first and the second end of the pipe are connected with each other,
vector representations of the ith layer user and kth class requirement object respectively are shown, when l is 0,
is Z->
Is O k P and W are weight parameters, wherein, indicates element-by-element multiplication operation, sigma is an activation function, and phi indicates that a vector is converted into a diagonal matrix;
s45: and repeating the operation in the S44, and alternately updating the vector representations of the user and the kth class of demand object respectively until the final layer of convolution is finished to obtain the vector representations of all the user and the kth class of demand object.
S5: aggregating neighbor information of a target user according to user social relations in a heterogeneous information network to obtain an expression vector of the target user and perform demand prediction, specifically comprising
S51: s4, obtaining a vector of the target user i and expressing the vector as
And a neighbor user j of the user i belongs to N (i), and the final target user vector representation is obtained by aggregating neighbor user information by using an attention mechanism. The weight coefficient of the neighbor to the target user is calculated,
the vector representation of the target user is updated,
wherein, α and W are weight parameters, σ is an activation function, and | is a splicing operation.
S52: for each target user, calculating its relevance prediction score to each kth class demand object
S53: the loss function is a binary cross entropy function:
wherein Y and Y - Positive and negative examples in the data set, Y represents the demand object set used by the user, Y - Sampled from unused demand objects in the data set by the user,
indicates whether the user has an interaction, presence interaction @, with the demand object>
Is 1, otherwise is 0. And (4) performing optimization solution on the loss function by using a random gradient descent method, sequencing the kth demand objects from high to low according to the prediction score obtained by S52 calculation, and selecting the first n demand objects as a kth demand list of the user.
S54: by repeating the operations of S42-S53, a list of all the categories of demand objects for each user can be obtained, thereby realizing active prediction of user demands.
The embodiment also provides a crowd-sourced user demand active prediction system, which comprises
A crowdsourcing task issuing module: determining annotators participating in the crowdsourcing task, designing the crowdsourcing task and issuing the crowdsourcing task to a crowdsourcing task platform, wherein the annotators receive the crowdsourcing task and complete the task; in particular to
S11: acquiring a user set of a service provider as a target user, acquiring user social relationship data from Twitter, a service provider platform and the like to obtain a social neighbor user set of the target user, and taking all users as annotators for receiving crowdsourcing tasks;
s12: a user preference survey questionnaire is designed from the aspects of demographic information, social requirements reflecting common interest in social relations, enjoyment requirements reflecting personal preferences and the like, the content of the questionnaire comprises word expressions and selection questions displayed in a graphical mode (for example, the interested content is selected from the following options), and a annotator is allowed to submit auxiliary information such as texts, videos, pictures and the like independently. And issuing the questionnaire to a crowdsourcing platform in a crowdsourcing task form, and issuing a task to the annotator acquired in S11.
Heterogeneous information network construction module: constructing a heterogeneous information network according to the user social relationship, the historical behavior data and the user preference information acquired in the crowdsourcing mode in the heterogeneous information network; in particular to
S21: taking the multi-modal data collected by S1 based on the crowdsourcing mode as a property set of each user, such as age, gender, favorite movie posters and the like, and properties of demand objects, such as manufacturers, dates of marketing and the like. The user and demand objects (including goods, services, etc.) are taken as nodes.
S22: establishing a connecting edge between the user and the demand object according to the following relation:
the relationship R1: direct relationships such as friends and concern exist among users U, and L are used respectively -1 Representing relationships between users U, i.e.
And &>
Relationship R2: some users have historical behavior information, such as the user bought a certain article, used a certain service, etc., respectively using B and B -1 Representing user U and requirement object O k In relation to each other, i.e.
And &>
Wherein k represents a kth class requirement object;
s23: and establishing a multi-mode heterogeneous information network according to the attribute set, the nodes and the relationship among the nodes.
The user requirement data space generation module: uniformly representing different types of entities in a heterogeneous information network to generate a user demand data space; in particular to
S31: user information and attribute information of a demand object collected based on a crowdsourcing mode generally have different expression forms, including types such as texts and pictures, and need to be uniformly expressed by adopting different expression learning methods according to different modalities:
obtaining vector representation of text type information by adopting word2vec method
Wherein e is u Representing a user text attribute vector representation, e o Representing the text attribute vector representation of the demand object, wherein N is the quantity of the demand object categories;
the picture type information is represented by vector obtained by adopting convolutional neural network
Wherein, g u Representing user picture attribute vector representation, g o A picture attribute vector representation representing a demand object;
s32: in order to learn the embedded representation of the nodes, the multimodality attribute information after being uniformly expressed needs to be fused. The user attribute vector e obtained in S31 is used u And g u Performing outer product operation to realize feature crossing, flattening the obtained matrix according to rows, and inputting the flattened matrix into a multilayer perceptron to obtain a user node initial vector representation Z and a demand object attribute representation vector
And &>
Repeating the operation to obtain the initial node of the demand objectVector representation O k The vector representations of all users and demand objects form a user demand data space.
The user and demand object representation vector learning module: extracting the interactive semantics of the user and the demand object by using a meta path in a heterogeneous information network, and respectively learning the expression vectors of the user and the demand object; in particular to
S41: establishing a plurality of user-demand object co-occurrence matrixes T according to historical behavior information of users k : for example, a user-item co-occurrence matrix
Subscriber-service co-occurrence matrix ≥>
Wherein, | I | is the quantity of articles, | S | is the quantity of services, if the user has bought a certain article or the user has used a certain service, put 1 in the corresponding position of the corresponding co-occurrence matrix;
s42: in the k-th demand active prediction scene of the user, extracting the UO from the heterogeneous information network constructed in the step S2 k U-element path, meaning that two users use the semantic information of the kth class demand object together, co-occurrence matrix T k To which it is transferred
Multiply, i.e. [ means ] in>
Get the inter-user relationship matrix->
Extracting O from the heterogeneous information network constructed in step S2 k UO k Meta-path representing semantic information used by the same user for two kth class demand objects by &>
Obtaining a relation matrix between kth class demand objects under the semantics/>
S43: for the relationship matrix between users obtained in S42
And a relationship matrix between demand objects>
The standardization treatment is carried out according to the following formulas respectively,
wherein the content of the first and second substances,
and &>
Are all diagonal matrices, are asserted>
And &>
Are respectively based on>
And &>
A degree matrix of (c);
s44: using a graph-convolution neural network, a user vector representation is learned according to the following formula,
the vector representation of the kth class demand object is learned in accordance with the following formula,
wherein, the first and the second end of the pipe are connected with each other,
vector representations of the l-th layer user and the k-th type demand object respectively are shown, when l is 0,
is Z->
Is O k P and W are weight parameters, wherein, indicates element-by-element multiplication operation, sigma is an activation function, and phi indicates that a vector is converted into a diagonal matrix;
s45: and repeating the operation in the S44, and alternately updating the vector representations of the user and the kth class of demand object respectively until the final layer of convolution is finished to obtain the vector representations of all the user and the kth class of demand object.
The demand prediction module is used for aggregating neighbor information of the target user according to the social relations of the users in the heterogeneous information network to obtain an expression vector of the target user and performing demand prediction, and specifically comprises
S51: the expression vector learning module of the user and the demand object obtains the vector expression of the target user i as
A neighbor user j of the user i belongs to N (i), and the final target user vector representation is obtained by aggregating neighbor user information by using an attention mechanism; the weight coefficient of the neighbor to the target user is calculated,
the vector representation of the target user is updated,
wherein alpha and W are weight parameters, sigma is an activation function, and | is splicing operation;
s52: for each target user, calculating the relevance prediction score of the target user and each k-th class demand object
S53: the loss function is a binary cross entropy function:
wherein Y and Y - Positive and negative examples in the data set, Y represents the demand object set used by the user, Y - Sampled from the demand objects in the data set that are not used by the user,
indicates whether the user has an interaction, presence interaction @, with the demand object>
Is 1, otherwise is 0; optimizing and solving the loss function by using a random gradient descent method, sequencing kth demand objects from high to low according to the prediction score calculated in the step S52, and selecting the first n demand objects as a kth demand list of the user;
s54: by repeating the operations of S42-S53, a list of all the categories of demand objects for each user can be obtained, thereby realizing the active prediction of the demand of the user.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (8)
1. A user demand active prediction method based on crowdsourcing is characterized by comprising the following steps: the method comprises the following steps:
s1: determining a annotator participating in the crowdsourcing task, designing the crowdsourcing task and issuing the crowdsourcing task to a crowdsourcing task platform, and receiving the crowdsourcing task and completing the task by the annotator; the method specifically comprises the following steps:
s11: acquiring a user set of a service provider as a target user, acquiring user social relationship data from a social network and a service provider platform to obtain a social neighbor user set of the target user, and taking all the users as annotators for receiving crowdsourcing tasks;
s12: designing a user preference survey questionnaire from the perspectives of demographic information, social requirements for reflecting common interest and love in social relations and enjoyment requirements for reflecting personal preferences, wherein the content of the questionnaire comprises word expression and selection questions displayed graphically, allowing a annotator to submit auxiliary information independently, publishing the questionnaire to a crowdsourcing platform in a crowdsourcing task form, and publishing tasks to the annotator obtained in S11;
s2: constructing a heterogeneous information network according to the user social relationship, the historical behavior data and the user preference information acquired in the crowdsourcing mode; the method specifically comprises the following steps:
s21: taking multi-modal data acquired based on a crowdsourcing mode in the S1 as an attribute set of each user, and taking the user and a demand object as nodes;
s22: establishing a connection edge between a user and a demand object according to the following relation:
the relationship R1: direct relationships of friends and concerns exist among users U, and L are used respectively -1 Representing relationships between users U, i.e.
And &>
Relationship R2: part of the users have historical behavior information, respectively B and B -1 Representing user U and demand object O k In relation to each other, i.e.
And &>
Wherein k represents a kth class requirement object;
s23: establishing a multi-mode heterogeneous information network according to the attribute set, the nodes and the relationship among the nodes;
s3: uniformly representing different types of entities in a heterogeneous information network to generate a user demand data space;
s4: extracting the interactive semantics of the user and the demand object by using a meta path in a heterogeneous information network, and respectively learning the expression vectors of the user and the demand object;
s5: and aggregating the neighbor information of the target user according to the social relation of the users in the heterogeneous information network to obtain the expression vector of the target user and perform demand prediction.
2. The crowd-sourced user demand active prediction method of claim 1, wherein: the step S3 includes:
s31: user information, text attribute information and image attribute information of a demand object collected in a crowdsourcing mode are uniformly expressed:
obtaining vector representation of text type information by adopting word2vec method
Wherein e is u Representing a user text attribute vector representation, e o Representing the text attribute vector representation of the demand object, wherein N is the quantity of the demand object categories;
representing vectors for the attributes of the demand objects;
the picture type information is represented by vector obtained by adopting convolutional neural network
Wherein, g u Representing user Picture Attribute vector representation, g o A picture attribute vector representation representing a demand object;
representing vectors for the attributes of the demand objects; />
S32: fusing the multi-mode attribute information after uniform expression: the user attribute vector e obtained in S31 is used u And g u Performing outer product operation to realize feature crossing, flattening the obtained matrix according to rows, and inputting the flattened matrix into a multilayer perceptron to obtain a user node initial vector representation Z and a demand object attribute representation vector
And &>
Repeating the operation to obtain the initial vector representation O of the demand object node k The vector representations of all users and demand objects form a user demand data space.
3. The crowd-sourced user demand active prediction method of claim 1, wherein: the step S4 includes:
s41: establishing a co-occurrence matrix T of a plurality of user-demand objects according to historical behavior information of users k : user-item co-occurrence matrix
Subscriber-service co-occurrence matrix ≥>
Wherein, | I | is the quantity of articles, | S | is the quantity of services, if the user has bought a certain article or the user has used a certain service, put 1 in the corresponding position of the corresponding co-occurrence matrix;
s42: in the k-th demand active prediction scene of the user, extracting the UO from the heterogeneous information network constructed in the step S2 k U-element path, meaning that two users use the semantic information of the kth class demand object together, co-occurrence matrix T k With its transpose (T) k ) T Multiplication, i.e. T k ×(T k ) T Obtaining the relationship matrix between the users under the semantic meaning
Extracting O from the heterogeneous information network constructed in step S2 k UO k Meta-path, meaning semantic information used by the same user for two kth class requirement objects, passing through (T) k ) T ×T k Obtain a relationship matrix ^ between the kth class of demand objects under the semantic>
S43: for the relationship matrix between users obtained in S42
And a relation matrix between demand objects>
The standardization treatment is carried out according to the following formulas respectively,
wherein, the first and the second end of the pipe are connected with each other,
and &>
Are all diagonal matrices, in combination>
And &>
Are respectively based on>
And &>
A degree matrix of (c);
s44: using a graph-convolution neural network, a user vector representation is learned according to the following formula,
the vector representation of the kth class demand object is learned in accordance with the following formula,
wherein the content of the first and second substances,
vector representations of the ith layer user and kth class demand object, respectively, are based on the fact that when l is 0, then->
Is a group of a group Z having a structure,
is O k P and W are weight parameters, wherein, indicates element-by-element multiplication operation, sigma is an activation function, and phi indicates that a vector is converted into a diagonal matrix;
s45: and repeating the operation in the S44, and alternately updating the vector representations of the user and the kth type demand object respectively until the last layer of convolution is finished to obtain the vector representations of all the users and the kth type demand object.
4. The active prediction method of user demand based on crowdsourcing of claim 3, wherein: the step S5 includes:
s51: s4, obtaining a vector of the target user i and expressing the vector as
Neighbor users j of the user i belong to N (i), and the neighbor user information is aggregated by using an attention mechanism to obtain final target user vector representation; the weight coefficient of the neighbor to the target user is calculated,
the vector representation of the target user is updated,
wherein, alpha and W are weight parameters, sigma is an activation function, and | l is splicing operation;
s52: for each target user, calculating its relevance prediction score to each kth class demand object
S53: the loss function is a binary cross entropy function:
wherein, Y and Y - Positive and negative examples in the data set, Y represents the demand object set used by the user, Y - Sampled from the demand objects in the data set that are not used by the user,
indicates whether a user has interacted with the demand object, there is an interaction @>
Is 1, otherwise is 0; optimizing and solving the loss function by using a random gradient descent method, sequencing kth-class demand objects from high to low according to the prediction score obtained by calculation in the step S52, and selecting the first n demand objects as a kth-class demand list of the user;
s54: by repeating the operations of S42-S53, a list of all the categories of demand objects for each user can be obtained, thereby realizing active prediction of user demands.
5. A crowd-sourced based active user demand prediction system is characterized in that: comprises that
A crowdsourcing task issuing module: determining a annotator participating in the crowdsourcing task, designing the crowdsourcing task and issuing the crowdsourcing task to a crowdsourcing task platform, and receiving the crowdsourcing task and completing the task by the annotator;
the specific execution process of the crowdsourcing task issuing module is as follows:
s11: acquiring a user set of a service provider as a target user, acquiring user social relationship data from a social network and a service provider platform to obtain a social neighbor user set of the target user, and taking all users as annotators for receiving crowdsourcing tasks;
s12: designing a user preference survey questionnaire from the aspects of demographic information, social requirements for reflecting common interest in social relations and enjoyment requirements for reflecting individual preferences, wherein the content of the questionnaire comprises selection questions expressed in characters and displayed in a graphical mode, a annotator is allowed to submit auxiliary information independently, the questionnaire is issued to a crowdsourcing platform in a crowdsourcing task mode, and tasks are issued to the annotator acquired in S11;
the heterogeneous information network construction module: constructing a heterogeneous information network according to the user social relationship, the historical behavior data and the user preference information acquired in the crowdsourcing mode;
the heterogeneous information network construction module comprises the following execution processes:
s21: taking multimodal data collected based on a crowdsourcing mode as an attribute set of each user, and taking the user and a demand object as nodes;
s22: establishing a connection edge between a user and a demand object according to the following relation:
the relationship R1: direct relationships such as friends and concern exist among users U, and L are used respectively -1 Representing relationships between users U, i.e.
And &>
The relationship R2: some users have historical behavior information, such as the user bought a certain article, used a certain service, etc., respectively using B and B -1 Representing user U and requirement object O k In relation to each other, i.e.
And &>
Wherein k represents a kth class requirement object;
s23: establishing a multi-mode heterogeneous information network according to the attribute set, the nodes and the relationship among the nodes;
the user requirement data space generation module: uniformly representing different types of entities in a heterogeneous information network to generate a user demand data space;
the user and demand object representation vector learning module: extracting the interactive semantics of the user and the demand object by using a meta path in the heterogeneous information network, and respectively learning the expression vectors of the user and the demand object;
a demand forecasting module: and aggregating the neighbor information of the target user according to the user social relationship in the heterogeneous information network to obtain the expression vector of the target user and perform demand prediction.
6. The crowd-sourced, user-demand active prediction system of claim 5, wherein: the specific execution process of the user requirement data space generation module comprises the following steps:
s31: user information, text attribute information and image attribute information of a demand object collected in a crowdsourcing mode are uniformly expressed:
obtaining vector representation of text type information by adopting word2vec method
Wherein e is u Representing a userText attribute vector representation, e o Representing the text attribute vector of the demand object, wherein N is the quantity of the demand object categories;
the picture type information is represented by vector obtained by adopting convolutional neural network
Wherein, g u Representing user picture attribute vector representation, g o A picture attribute vector representation representing a demand object;
s32: fusing the multi-mode attribute information after uniform expression: the user attribute vector e obtained in S31 is used u And g u Performing outer product operation to realize feature intersection, flattening the obtained matrix according to rows, inputting the flattened matrix into a multilayer perceptron to obtain an initial vector representation Z of a user node, and expressing a vector of a demand object attribute
And &>
Repeating the operation to obtain the initial vector representation O of the demand object node k The vector representations of all users and demand objects form a user demand data space.
7. The crowd-sourced, user-demand active prediction system of claim 6, wherein: the specific implementation process of the expression vector learning module of the user and the demand object comprises the following steps:
s41: establishing a co-occurrence matrix T of a plurality of user-demand objects according to historical behavior information of users k : user-item co-occurrence matrix
Subscriber-service co-occurrence matrix ≥>
Wherein, | I | is the quantity of articles, | S | is the quantity of services, if the user has bought a certain article or the user has used a certain service, put 1 in the corresponding position of the corresponding co-occurrence matrix;
s42: in the k-th demand active prediction scene of the user, extracting the UO from the constructed heterogeneous information network k U-element path, meaning that two users use the semantic information of the kth class demand object together, co-occurrence matrix T k With its transpose (T) k ) T Multiplication, i.e. T k ×(T k ) T Obtaining the relationship matrix between the users under the semantic meaning
Extracting O from a constructed heterogeneous information network k UO k Meta-path, meaning semantic information used by the same user for two kth class requirement objects, passing through (T) k ) T ×T k Obtain a relationship matrix ^ between the kth class of demand objects under the semantic>
S43: for the relationship matrix between users obtained in S42
And a relation matrix between demand objects>
The standardization treatment is carried out according to the following formulas respectively,
wherein the content of the first and second substances,
and &>
Are all diagonal matrices, are asserted>
And &>
Are respectively in>
And &>
A degree matrix of (c);
s44: using a graph convolution neural network, a user vector representation is learned in accordance with the following formula,
the vector representation of the kth class demand object is learned as follows,
wherein, the first and the second end of the pipe are connected with each other,
vector representations of the ith layer user and kth class demand object, respectively, are based on the fact that when l is 0, then->
Is a group of Z and is a group of Z,
is O k P and W are weight parameters, wherein, indicates element-by-element multiplication operation, sigma is an activation function, and phi indicates that a vector is converted into a diagonal matrix;
s45: and repeating the operation in the S44, and alternately updating the vector representations of the user and the kth class of demand object respectively until the final layer of convolution is finished to obtain the vector representations of all the user and the kth class of demand object.
8. The crowd-sourced, user-demand active prediction system of claim 7, wherein: the specific execution process of the step demand forecasting module comprises the following steps:
s51: the expression vector learning module of the user and the demand object obtains the vector expression of the target user i as
A neighbor user j of the user i belongs to N (i), and the final target user vector representation is obtained by aggregating neighbor user information by using an attention mechanism; the weight coefficient of the neighbor to the target user is calculated,
the vector representation of the target user is updated,
wherein, alpha and W are weight parameters, sigma is an activation function, and | l is splicing operation;
s52: for each target user, calculating its relevance prediction score to each kth class demand object
S53: the loss function is a binary cross entropy function:
wherein Y and Y - Positive and negative examples in the data set, Y represents the demand object set used by the user, Y - Sampled from the demand objects in the data set that are not used by the user,
indicates whether the user has an interaction, presence interaction @, with the demand object>
Is 1, otherwise is 0; optimizing and solving the loss function by using a random gradient descent method, sequencing kth demand objects from high to low according to the prediction score calculated in the step S52, and selecting the first n demand objects as a kth demand list of the user;
s54: by repeating the operations of S42-S53, a list of all the categories of demand objects for each user can be obtained, thereby realizing active prediction of user demands.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011387991.5A CN112508256B (en) | 2020-12-01 | 2020-12-01 | User demand active prediction method and system based on crowdsourcing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011387991.5A CN112508256B (en) | 2020-12-01 | 2020-12-01 | User demand active prediction method and system based on crowdsourcing |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112508256A CN112508256A (en) | 2021-03-16 |
| CN112508256B true CN112508256B (en) | 2023-04-14 |
Family
ID=74969199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011387991.5A Active CN112508256B (en) | 2020-12-01 | 2020-12-01 | User demand active prediction method and system based on crowdsourcing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112508256B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113361928B (en) * | 2021-06-07 | 2023-08-25 | 南京大学 | Crowd-sourced task recommendation method based on heterogram attention network |
| CN113378051B (en) * | 2021-06-16 | 2024-03-22 | 南京大学 | User-task association crowdsourcing task recommendation method based on graph neural network |
| CN113393056B (en) * | 2021-07-08 | 2022-11-25 | 山东大学 | Crowdsourcing service supply and demand gap prediction method and system based on time sequence |
| CN114445043B (en) * | 2022-01-26 | 2022-12-16 | 安徽大学 | Open ecological cloud ERP-based heterogeneous graph user demand accurate discovery method and system |
| CN114470790A (en) * | 2022-02-09 | 2022-05-13 | 腾讯科技（深圳）有限公司 | Virtual resource processing method, device, equipment, computer program and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191081A (en) * | 2019-12-17 | 2020-05-22 | 安徽大学 | Developer recommendation method and device based on heterogeneous information network |
| CN111626616A (en) * | 2020-05-27 | 2020-09-04 | 深圳莫比嗨客数据智能科技有限公司 | Crowdsourcing task recommendation method |
| CN111881342A (en) * | 2020-06-23 | 2020-11-03 | 北京工业大学 | Recommendation method based on graph twin network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11463472B2 (en) * | 2018-10-24 | 2022-10-04 | Nec Corporation | Unknown malicious program behavior detection using a graph neural network |
-
2020
- 2020-12-01 CN CN202011387991.5A patent/CN112508256B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191081A (en) * | 2019-12-17 | 2020-05-22 | 安徽大学 | Developer recommendation method and device based on heterogeneous information network |
| CN111626616A (en) * | 2020-05-27 | 2020-09-04 | 深圳莫比嗨客数据智能科技有限公司 | Crowdsourcing task recommendation method |
| CN111881342A (en) * | 2020-06-23 | 2020-11-03 | 北京工业大学 | Recommendation method based on graph twin network |
Non-Patent Citations (3)
| Title |
|---|
| "Heterogeneous Information Network Embedding with Convolutional Graph Attention Networks";Meng Cao等;《2020 International Joint Conference on Neural Networks (IJCNN)》;IEEE;20200928;第2020年卷;第1-8页 * |
| "基于异质信息网络表示学习的推荐算法研究与实现";胡斌斌;《中国优秀硕士学位论文全文数据库信息科技辑》;20190815;第2019年卷(第08期);第I138-1361页 * |
| 基于融合元路径图卷积的异质网络表示学习;蒋宗礼等;《计算机科学》;20200731;第47卷(第07期);第231-235页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112508256A (en) | 2021-03-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112508256B (en) | User demand active prediction method and system based on crowdsourcing | |
| Bi et al. | Modelling customer satisfaction from online reviews using ensemble neural network and effect-based Kano model | |
| CN108431833B (en) | End-to-end depth collaborative filtering | |
| US11580447B1 (en) | Shared per content provider prediction models | |
| US10102503B2 (en) | Scalable response prediction using personalized recommendation models | |
| Lytvyn et al. | Design of the architecture of an intelligent system for distributing commercial content in the internet space based on SEO-technologies, neural networks, and Machine Learning | |
| EP4181026A1 (en) | Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium | |
| US20230259964A1 (en) | Device for providing mediation service between advertiser and influencer by using artificial intelligence, and mediation method using same | |
| CN111241394B (en) | Data processing method, data processing device, computer readable storage medium and electronic equipment | |
| Khalid et al. | A literature review of implemented recommendation techniques used in massive open online courses | |
| Kutlimuratov et al. | Modeling and applying implicit dormant features for recommendation via clustering and deep factorization | |
| CN111429161B (en) | Feature extraction method, feature extraction device, storage medium and electronic equipment | |
| Shen et al. | A voice of the customer real-time strategy: An integrated quality function deployment approach | |
| Wu et al. | Enhanced e-commerce customer engagement: A comprehensive three-tiered recommendation system | |
| Wang | A survey of online advertising click-through rate prediction models | |
| Wang et al. | Who are the best adopters? User selection model for free trial item promotion | |
| CN113610610B (en) | Session recommendation method and system based on graph neural network and comment similarity | |
| KR102238438B1 (en) | System for providing commercial product transaction service using price standardization | |
| US20230316106A1 (en) | Method and apparatus for training content recommendation model, device, and storage medium | |
| Vysotska et al. | V. Lytvyn | |
| Yu et al. | A graph attention network under probabilistic linguistic environment based on Bi-LSTM applied to film classification | |
| CN114090848A (en) | Data recommendation and classification method, feature fusion model and electronic equipment | |
| Lakshmi Chetana et al. | CF-AMVRGO: Collaborative Filtering based Adaptive Moment Variance Reduction Gradient Optimizer for Movie Recommendations | |
| CN111460300A (en) | Network content pushing method and device and storage medium | |
| Gao et al. | DDRCN: Deep Deterministic Policy Gradient Recommendation Framework Fused with Deep Cross Networks |
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 |