CN113781110B - User behavior prediction method and system based on multi-factor weighted BI-LSTM learning - Google Patents
User behavior prediction method and system based on multi-factor weighted BI-LSTM learning Download PDFInfo
- Publication number
- CN113781110B CN113781110B CN202111042979.5A CN202111042979A CN113781110B CN 113781110 B CN113781110 B CN 113781110B CN 202111042979 A CN202111042979 A CN 202111042979A CN 113781110 B CN113781110 B CN 113781110B
- Authority
- CN
- China
- Prior art keywords
- behavior
- user
- target user
- node
- lstm
- 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
- 230000006399 behavior Effects 0.000 title claims abstract description 207
- 238000000034 method Methods 0.000 title claims abstract description 29
- 239000013598 vector Substances 0.000 claims abstract description 55
- 230000007246 mechanism Effects 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 6
- 238000013519 translation Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 4
- 230000003542 behavioural effect Effects 0.000 claims description 3
- 238000007405 data analysis Methods 0.000 abstract description 2
- 238000013480 data collection Methods 0.000 abstract description 2
- 230000015654 memory Effects 0.000 description 4
- 230000002457 bidirectional effect Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000003631 expected effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Accounting & Taxation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Finance (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Development Economics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Entrepreneurship & Innovation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- Probability & Statistics with Applications (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a user behavior prediction method based on multi-factor weighted BI-LSTM learning, which comprises the following steps: converting the collected historical behavior description text of the target user into a vector representation form by Word2Vec, and extracting the historical behavior characteristics of the target user from the historical behavior description text by using a transducer; according to the historical behavior characteristics of the target user, self-adaptively learning multi-factor weights of the historical behaviors based on an attention mechanism; and according to the multi-factor weight, performing behavior prediction by utilizing the BI-LSTM and the bundle searching strategy. According to the invention, the behavior characteristics of the user are learned by directly analyzing the user history behavior description text and the known social network, and compared with the traditional method of capturing the user behavior from the data and then predicting the behavior, the data collection method is more flexible, the data analysis method is more various, and the universality is better. The invention also provides a corresponding user behavior prediction system based on multi-factor weighted BI-LSTM learning.
Description
Technical Field
The invention belongs to the technical field of behavior prediction, and particularly relates to a user behavior prediction method and system based on multi-factor weighted BI-LSTM learning.
Background
When recommending commodities to users on an electronic commerce live broadcast platform, behavior research of target users is often influenced by multiple factors in the prediction process of evaluating the behaviors (clicking, purchasing, collecting and the like) of the users, and especially social influence of the users is ignored, so that the behaviors of the users are difficult to predict.
Many current studies predict the behavior of users based on their influence in social networks, most of which have problems of distortion of influence assessment and no comprehensive multifactor prediction, resulting in poor expected effect of the user behavior. The behavior of users on an e-commerce live platform is affected by a variety of factors, especially behavior factors of other users with the same preferences and social relationships. Only if the influence factors are integrated to carry out global judgment, a more accurate prediction result can be obtained.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the target user behavior prediction method provided by the invention realizes analysis of the change rule and the development trend of the target user behavior, and compared with the prior art, the behavior prediction method based on multi-factor weighted bidirectional long-short-Term Memory (Bi-directional Long Short-Term Memory, BI-LSTM) learning firstly utilizes a tranformer translation model of google to extract the global semantic features of a target user historical behavior description text in a deep level, then extracts influence factors through a social knowledge graph of the user, and finally predicts the behavior based on the trained weighted BI-LSTM, thereby being convenient for further and more efficient recommendation.
To achieve the above object, according to one aspect of the present invention, there is provided a user behavior prediction method based on multi-factor weighted BI-LSTM learning, comprising:
step S1: converting the collected historical behavior description text of the target user into a vector representation form by Word2Vec, and extracting the historical behavior characteristics of the target user from the historical behavior description text by using a transducer;
step S2: according to the historical behavior characteristics of the target user, self-adaptively learning multi-factor weights of the historical behaviors based on an attention mechanism;
step S3: and according to the multi-factor weight, performing behavior prediction by utilizing the BI-LSTM and the bundle searching strategy.
In one embodiment of the present invention, the step S1 specifically includes: firstly, based on collected target user history behavior description text, firstly, word segmentation is carried out on the target user history behavior description text, then each Word is converted into a vector representation form through a Word vector model Word2Vec, and an initialized text vector matrix is constructedEach of which is +.>The embedded vector representing a word IN a sentence is input into a transducer translation model, a self-attention mechanism self-attention is continuously learned to obtain a feature matrix IA of a target user history behavior description text, and each behavior of the matrix IA is a feature vector of a word.
In one embodiment of the present invention, the multi-factor weights in step S2 include: current user behavior impact w 1 And other user behavior influences w 2 。
In one embodiment of the invention, the current user behavior affects w 1 The calculation mode of (a) is as follows:
wherein,represents the kth shortest path between nodes, and assumes a total of t shortest paths, PI, between two target user nodes u,v Representing the propagation of the behavioral impact between two target user nodes.
In one embodiment of the invention, the behavior between the two target user nodes affects propagation PI u,v Calculated using the following formula:
wherein,represents the kth path between node u and node v,>representing node v i And node v j The relation weight between (v) i ,v j ) Representing nodes on all paths between node u and node v, including node u and nodes v, F i Representing each user node v i Behavior influence of (c) is provided.
In one embodiment of the invention, the user node v i Behavior influence F of (2) i Calculated using the following formula:
F i ＝tr(IA(i))
tr (IA (i)) represents a user node v i The corresponding user history behavior describes the trace of the feature matrix IA of the text.
In one embodiment of the present invention,the other user behavior influences w 2 The method comprises the following steps: w (w) 2 ＝defree(v i ) Degree computation using graph theory nodes.
In one embodiment of the present invention, the step S3 specifically includes: taking a characteristic matrix IA formed by historical behavior description texts of each target user as input of a BI-LSTM network, training the BI-LSTM to learn behavior characteristics of each target user, and outputting probability values of the user on article behaviors under the condition of known target users and commodities, wherein the probability vector output by the forward LSTM isThe probability vector of the backward LSTM output is +.>The probability values of three types of behaviors in the behavior space are recorded by two probability vectors, namely, the types of behaviors captured in the user history behavior description text are added into model training for a target user v by the influence weights of the target user and other users i Obtain a weighted output probability vector +.>
In one embodiment of the present invention, the three types of behaviors in the behavior space specifically include: clicking, purchasing and collecting.
According to another aspect of the present invention, there is also provided a system for predicting user behavior based on multi-factor weighted BI-LSTM learning, including a target user history behavior feature extraction module, a multi-factor weight learning module, and a behavior prediction module, wherein:
the target user history behavior feature extraction module is used for converting the collected target user history behavior description text into a vector representation form from Word2Vec, and then extracting target user history behavior features from the user history behavior description text by using a transformer;
the multi-factor weight learning module is used for adaptively learning the multi-factor weight of the historical behavior based on the attention mechanism according to the historical behavior characteristics of the target user;
the behavior prediction module is used for performing behavior prediction by utilizing the BI-LSTM and the cluster search strategy according to the multi-factor weight.
In general, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects:
(1) The user behavior characteristics are learned by directly analyzing the user history behavior description text and the known social network, so that compared with the traditional method that the user behavior is firstly captured from the data and then the behavior prediction is carried out, the data collection method is more flexible, the data analysis method is more various, and the universality is better;
(2) According to the historical behavior description text of the target user and social network information, the behavior influence among users is considered in the two directions of the sent influence and the received influence, and compared with the traditional method of integrally considering influence propagation in the network, the social influence situation of the users in real life is more closely considered;
(3) The user history behavior description text and behavior prediction are linked by training the bidirectional long-short-time memory network, the problems that the text information is underutilized, the text information is lost, the disturbance caused by irrelevant information is large and the like in the traditional memory network method can be solved by the bidirectional long-short-time memory network, and the accuracy of behavior prediction can be improved by adding the weight factor for adjustment.
Drawings
FIG. 1 is a flow chart of a behavior prediction method based on multi-factor BI-LSTM learning in an embodiment of the invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
As shown in FIG. 1, the invention provides a behavior prediction method based on multi-factor BI-LSTM learning, which comprises the following steps:
step S1: converting the collected historical behavior description text of the target user into a vector representation form by Word2Vec, and extracting the historical behavior characteristics of the target user from the historical behavior description text by using a transducer;
specifically, the step S1 includes:
firstly, based on collected target user history behavior description text, firstly, word segmentation is carried out on the target user history behavior description text, then each Word is converted into a vector representation form through a Word vector model Word2Vec, and an initialized text vector matrix is constructedEach of which is +.>An embedded vector representing a word in a sentence. The vector matrix IN is input into a transducer translation model, and a self-attention mechanism self-attention is continuously learned to obtain a feature matrix IA of the target user history behavior description text, wherein each behavior of the matrix IA is a feature vector of a word.
Step S2: according to the historical behavior characteristics of the target user, self-adaptively learning multi-factor weights of the historical behaviors based on an attention mechanism;
specifically, the current user behavior influence w is to be calculated in said step S2 1 Influence of other user behavior w 2 . The calculation method is as follows:
step S21: current user behavior impact w based on shortest maximum propagation policy 1 And (5) calculating. w (w) 1 The initial value is calculated as follows:
the target user knowledge graph is represented in the form of a graph, i.e., g= (V, E, W), where V is a set of target user nodes, E represents a set of relationships between the target user nodes, and W represents a weight matrix of the relationships between the target user nodes. By means of shortest-most propagation strategy, i.e. selecting target user nodesThe shortest propagation path between points and its behavior affects the propagation intensity the most. The relationship weight between the target user nodes is represented by w, and F represents the behavior influence of the target user nodes. Let P be u,v Representing the set of all paths for node u and node v, the behavior impact propagation between two target user nodes can be calculated by:
wherein,represents the kth path between node u and node v,>representing node v i And node v j The relation weight between (v) i ,v j ) Representing nodes on all paths between node u and node v, including node u and node v. F (F) i Representing each user node v i The behavior influence of (2) is calculated as follows:
F i ＝tr(IA(i))
tr (IA (i)) represents a user node v i The corresponding user history behavior describes the trace of the feature matrix IA of the text. Final current user behavior influence w 1 The calculation is as follows:
wherein,represents the kth shortest path between nodes and assumes a total of t shortest paths between two target user nodes.
Step S22: other user behavior influence w based on network structure 2 And (5) calculating.
For a targetUser v i In other words, the influence of other user behavior depends on the neighbor relation between users, and thus, the other user behavior influences w 2 ＝degree(v i ) Degree computation of graph theory nodes is used herein.
Step S3: and according to the multi-factor weight, performing behavior prediction by utilizing the BI-LSTM and the bundle searching strategy.
Specifically, the step S3 includes:
taking a characteristic matrix IA formed by the historical behavior description text of each target user as input of a BI-LSTM network, training the BI-LSTM to learn the behavior characteristics of each target user, and outputting probability values of the user on the behavior of the article under the condition of known target users and commodities. Wherein, the probability vector of forward LSTM output is recorded asThe probability vector of the backward LSTM output isThe probability values of three types of behaviors of the behavior space { click, purchase and collection } are recorded by the two probability vectors, namely, the behavior types captured in the user history behavior description text. Adding the influence weights of the target user and other users into model training for one target user v i Obtain a weighted output probability vector +.>
It should be noted that, the two vectors are vectors composed of three elements, where each element corresponds to a probability value of a behavior class, and only the behaviors of the three classes that are most common and have influence are considered in the method of the present invention. The trained weighted BI-LSTM can output a behavior probability prediction result for the user by inputting a historical description text vector matrix of the user.
The process according to the invention is further illustrated in the following in connection with the specific examples:
(1) Extracting historical behavior characteristics of a target user:
first baseFirstly, word segmentation is carried out on the collected target user history behavior description text, then each Word is converted into a vector representation form from Word2Vec, and an initialized text vector matrix is constructedEach of which is +.>(l=1, 2, … m) represents an embedded vector of one word in a sentence. The vector matrix IN is input into a transducer translation model, and a self-attention mechanism self-attention is continuously learned to obtain a feature matrix IA of the target user history behavior description text, wherein each behavior of the matrix IA is a feature vector of a word.
Taking "Zhang Sanhe film" as an example, firstly, the words of the "Zhang Sanhe film" are divided into "Zhang Sanhe" and "favorite" film "which are converted into Word vector matrix by Word2VecAfter inputting a transducer, learning to obtain a feature matrix +.>Wherein->Word vectors respectively representing Zhang Sang, favorite and filmHaving the same dimensions.
(2) Multi-factor weights for learning historical behavior adaptively based on an attention mechanism:
firstly, mining influence factors from a user behavior knowledge graph, wherein the invention selects two main influence factors for analysis, and the two main influence factors are respectively influence w for the current user behavior 1 Influence of other user behavior w 2 . The calculation method is as follows:
step S21: current user behavior impact w based on shortest maximum propagation policy 1 And (5) calculating.
w 1 The initial value is calculated as follows:
the target user knowledge graph is represented in the form of a graph, i.e., g= (V, E, W), where V is a set of target user nodes, E represents a set of relationships between the target user nodes, and W represents a weight matrix of the relationships between the nodes. The shortest maximum propagation strategy method is adopted, namely the shortest propagation path between target user nodes is selected, and the behavior of the shortest propagation path affects the maximum propagation intensity. The relationship weight between the target user nodes is represented by w, and F represents the behavior influence of the target user nodes. Let P be u,v Representing the set of all paths for node u and node v, the behavior impact propagation between two target user nodes can be calculated as follows:
wherein,represents the kth path between node u and node v,>representing node v i And node v j The relation weight between (v) i ,v j ) Representing nodes on all paths between node u and node v, including node u and nodes v, F i Representing each user node v i Behavior influence of (c) is provided. The calculation method is as follows:
F i ＝tr(IA(i))
tr (IA (i)) represents a user node v i The corresponding user history behavior describes the trace of the feature matrix IA of the text.
tr (IA (i)) represents a user node v i The corresponding user history behavior describes the feature matrix IA of the text. Final current user behavior influence w 1 The calculation is as follows:
wherein,represents the kth shortest path between nodes and assumes a total of t shortest paths between two target user nodes.
Step S22: other user behavior influence w based on network structure 2 And (5) calculating.
For a target user v i In other words, the influence of other user behavior depends on the neighbor relation between users, so other user behavior influences w 2 ＝degree(v i ) Degree computation of graph theory nodes is used herein.
(3) Behavior prediction using weighted BI-LSTM and bundle search strategy:
taking a characteristic matrix IA formed by the historical behavior description text of each target user as input of the BI-LSTM, training the BI-LSTM to learn the behavior characteristics of each target user, and outputting probability values of the user on the behavior of the article under the condition of known target users and commodities. Wherein, the probability vector of forward LSTM output is recorded asThe probability vector of the backward LSTM output is +.>The probability values of three types of behaviors of the behavior space { click, purchase and collection } are recorded by the two probability vectors, namely, the behavior types captured in the user history behavior description text. Adding the influence weights of the target user and other users into model training for one target user v i Obtain a weighted output probability vector +.>
The trained weighted BI-LSTM can output a behavior probability prediction result for the user by inputting a historical description text vector matrix of the user.
Further, the invention also provides a user behavior prediction system based on multi-factor weighted BI-LSTM learning, which comprises a target user history behavior feature extraction module, a multi-factor weight learning module and a behavior prediction module, wherein:
the target user history behavior feature extraction module is used for converting the collected target user history behavior description text into a vector representation form from Word2Vec, and then extracting target user history behavior features from the user history behavior description text by using a transformer;
the multi-factor weight learning module is used for adaptively learning the multi-factor weight of the historical behavior based on the attention mechanism according to the historical behavior characteristics of the target user;
the behavior prediction module is used for performing behavior prediction by utilizing the BI-LSTM and the cluster search strategy according to the multi-factor weight.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (4)
1. A method for predicting user behavior based on multi-factor weighted BI-LSTM learning, comprising:
step S1: converting the collected historical behavior description text of the target user into a vector representation form by Word2Vec, and extracting the historical behavior characteristics of the target user from the historical behavior description text of the target user by using a transducer;
step S2: according to the historical behavior characteristics of the target user, self-adaptively learning multi-factor weights of the historical behaviors based on an attention mechanism; wherein, the multi-factor weights in the step S2 include: current user behavior impact w 1 And other user behavior influences w 2 The method comprises the steps of carrying out a first treatment on the surface of the The current user behavior influences w 1 The calculation mode of (a) is that：Wherein->Represents the kth shortest path between nodes, and assumes a total of t shortest paths, PI, between two target user nodes u,v Representing behavior impact propagation between two target user nodes; propagation of PI by behavioral influences between the two target user nodes u,v Calculated using the following formula: />Wherein (1)>Represents the kth path between node u and node v,>representing node v i And node v j The relation weight between (v) i ,v j ) Representing nodes on all paths between node u and node v, including node u and nodes v, F i Representing each user node v i Behavior influence of (a) is determined; user node v i Behavior influence F of (2) i Calculated using the following formula: f (F) i Tr (IA (i)), tr (IA (i)) represents the user node v i Trace of feature matrix IA of corresponding user history behavior descriptive text; the other user behavior influences w 2 The method comprises the following steps: w (w) 2 ＝degree(v i ) Calculating the degree of the nodes by using the graph theory;
step S3: according to the multi-factor weight, performing behavior prediction by utilizing a BI-LSTM and a bundle searching strategy; the step S3 specifically includes: taking a feature matrix IA formed by historical behavior description texts of each target user as input of a BI-LSTM network, training the BI-LSTM to learn behavior features of each target user, and outputting the behavior of the user on the article under the condition of known target users and commoditiesProbability values, wherein the probability vector output by the forward LSTM is recorded asThe probability vector of the backward LSTM output isThe probability values of three types of behaviors in the behavior space, namely the types of behaviors captured in the user history behavior description text, are recorded by two probability vectors, and the current user behavior is influenced by w 1 And other user behavior influences w 2 To be added into model training, for a target user v i Obtain a weighted output probability vector +.>
2. The method for predicting user behavior based on multi-factor weighted BI-LSTM learning of claim 1, wherein said step S1 specifically comprises:
firstly, based on collected target user history behavior description text, firstly, word segmentation is carried out on the target user history behavior description text, then each Word is converted into a vector representation form through a Word vector model Word2Vec, and an initialized text vector matrix is constructedEach of which is +.>The embedded vector representing a word IN a sentence, i=1, 2, … m, the vector matrix IN is input into a transducer translation model, and the self-attention mechanism self-attention learning is continuously carried out to obtain a feature matrix IA of the target user history behavior description text, wherein each feature vector of the matrix IA acts as a word.
3. The method for predicting user behavior based on multi-factor weighted BI-LSTM learning of claim 1, wherein the three classes of behavior space specifically include: clicking, purchasing and collecting.
4. The system for predicting the user behavior based on multi-factor weighted BI-LSTM learning is characterized by comprising a target user history behavior feature extraction module, a multi-factor weight learning module and a behavior prediction module, wherein:
the target user history behavior feature extraction module is used for converting the collected target user history behavior description text into a vector representation form from Word2Vec, and then extracting target user history behavior features from the target user history behavior description text by using a transformer;
the multi-factor weight learning module is used for adaptively learning the multi-factor weight of the historical behavior based on the attention mechanism according to the historical behavior characteristics of the target user; wherein the multi-factor weights include: current user behavior impact w 1 And other user behavior influences w 2 The method comprises the steps of carrying out a first treatment on the surface of the The current user behavior influences w 1 The calculation mode of (a) is as follows:
wherein->Represents the kth shortest path between nodes, and assumes a total of t shortest paths, PI, between two target user nodes u,v Representing behavior impact propagation between two target user nodes; propagation of PI by behavioral influences between the two target user nodes u,v Calculated using the following formula: />Wherein (1)>Representing the kth path between node u and node v，/>Representing node v i And node v j The relation weight between (v) i ,v j ) Representing nodes on all paths between node u and node v, including node u and nodes v, F i Representing each user node v i Behavior influence of (a) is determined; user node v i Behavior influence F of (2) i Calculated using the following formula: f (F) i Tr (IA (i)), tr (IA (i)) represents the user node v i Trace of feature matrix IA of corresponding user history behavior descriptive text; the other user behavior influences w 2 The method comprises the following steps: w (w) 2 ＝degree(v i ) Calculating the degree of the nodes by using the graph theory;
the behavior prediction module is used for performing behavior prediction by utilizing a BI-LSTM and a bundle searching strategy according to the multi-factor weight; the step S3 specifically includes: taking a feature matrix IA formed by historical behavior description texts of each target user as input of a BI-LSTM network, training the BI-LSTM to learn behavior features of each target user, and outputting probability values of the user on the behavior of the object under the condition of known target users and commodities, wherein the probability vector output by the forward LSTM isThe probability vector of the backward LSTM output is +.>The probability values of three types of behaviors in the behavior space, namely the types of behaviors captured in the user history behavior description text, are recorded by two probability vectors, and the current user behavior is influenced by w 1 And other user behavior influences w 2 To be added into model training, for a target user v i Obtain a weighted output probability vector +.>
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111042979.5A CN113781110B (en) | 2021-09-07 | 2021-09-07 | User behavior prediction method and system based on multi-factor weighted BI-LSTM learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111042979.5A CN113781110B (en) | 2021-09-07 | 2021-09-07 | User behavior prediction method and system based on multi-factor weighted BI-LSTM learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113781110A CN113781110A (en) | 2021-12-10 |
| CN113781110B true CN113781110B (en) | 2024-03-26 |
Family
ID=78841417
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111042979.5A Active CN113781110B (en) | 2021-09-07 | 2021-09-07 | User behavior prediction method and system based on multi-factor weighted BI-LSTM learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113781110B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334843A (en) * | 2019-04-22 | 2019-10-15 | 山东大学 | A kind of time-varying attention improves be hospitalized medial demand prediction technique and the device of Bi-LSTM |
| CN111046294A (en) * | 2019-12-27 | 2020-04-21 | 支付宝(杭州)信息技术有限公司 | Click rate prediction method, recommendation method, model, device and equipment |
| CN111241421A (en) * | 2020-01-14 | 2020-06-05 | 西安电子科技大学 | User forwarding behavior prediction method based on social context information |
| CN112115358A (en) * | 2020-09-14 | 2020-12-22 | 中国船舶重工集团公司第七0九研究所 | Personalized recommendation method using multi-hop path features in knowledge graph |
| WO2021021330A1 (en) * | 2019-07-30 | 2021-02-04 | Intuit Inc. | Neural network system for text classification |
| CN112417314A (en) * | 2020-11-26 | 2021-02-26 | 清华大学 | Social network suicidal ideation detection method and system |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103294800B (en) * | 2013-05-27 | 2016-12-28 | 华为技术有限公司 | A kind of information-pushing method and device |
| CN110083705B (en) * | 2019-05-06 | 2021-11-02 | 电子科技大学 | Multi-hop attention depth model, method, storage medium and terminal for target emotion classification |
| US11954577B2 (en) * | 2019-09-13 | 2024-04-09 | Intuit Inc. | Deep neural network based user segmentation |
-
2021
- 2021-09-07 CN CN202111042979.5A patent/CN113781110B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110334843A (en) * | 2019-04-22 | 2019-10-15 | 山东大学 | A kind of time-varying attention improves be hospitalized medial demand prediction technique and the device of Bi-LSTM |
| WO2021021330A1 (en) * | 2019-07-30 | 2021-02-04 | Intuit Inc. | Neural network system for text classification |
| CN111046294A (en) * | 2019-12-27 | 2020-04-21 | 支付宝(杭州)信息技术有限公司 | Click rate prediction method, recommendation method, model, device and equipment |
| CN111241421A (en) * | 2020-01-14 | 2020-06-05 | 西安电子科技大学 | User forwarding behavior prediction method based on social context information |
| CN112115358A (en) * | 2020-09-14 | 2020-12-22 | 中国船舶重工集团公司第七0九研究所 | Personalized recommendation method using multi-hop path features in knowledge graph |
| CN112417314A (en) * | 2020-11-26 | 2021-02-26 | 清华大学 | Social network suicidal ideation detection method and system |
Non-Patent Citations (4)
| Title |
|---|
| 一种基于用户播放行为序列的个性化视频推荐策略;王娜;何晓明;刘志强;王文君;李霞;;计算机学报(01);125-137 * |
| 基于深度学习的词汇表示模型对比研究;余传明，王曼怡，林虹君，朱星宇，黄婷婷，安璐;数据分析与知识发现(第8期);28-40 * |
| 用于下一项推荐的序列感知深度网络;赵串串;游进国;李晓武;;小型微型计算机系统(07);47-52 * |
| 社交网络中用户行为分析研究;李新焕;赵应丁;;计算机时代(06);33-35+39 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113781110A (en) | 2021-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111061856B (en) | Knowledge perception-based news recommendation method | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| JP5454357B2 (en) | Information processing apparatus and method, and program | |
| CN112256866B (en) | Text fine-grained emotion analysis algorithm based on deep learning | |
| CN105447038A (en) | Method and system for acquiring user characteristics | |
| JP7393060B2 (en) | Personalized search method and search system combining attention mechanism | |
| CN113449204B (en) | Social event classification method and device based on local aggregation graph attention network | |
| Yu et al. | A novel weight-oriented graph convolutional network for aspect-based sentiment analysis | |
| Chen et al. | User-QoS-based web service clustering for QoS prediction | |
| CN114637911A (en) | Next interest point recommendation method of attention fusion perception network | |
| CN111753151B (en) | Service recommendation method based on Internet user behavior | |
| CN112380427B (en) | User interest prediction method based on iterative graph attention network and electronic device | |
| CN109344319B (en) | Online content popularity prediction method based on ensemble learning | |
| CN113781110B (en) | User behavior prediction method and system based on multi-factor weighted BI-LSTM learning | |
| CN112052995A (en) | Social network user influence prediction method based on fusion emotional tendency theme | |
| CN111523604A (en) | User classification method and related device | |
| Hao et al. | Deep collaborative online learning resource recommendation based on attention mechanism | |
| CN112270334B (en) | Few-sample image classification method and system based on abnormal point exposure | |
| CN114757391A (en) | Service quality prediction method based on network data space design | |
| CN114117229A (en) | Project recommendation method of graph neural network based on directed and undirected structural information | |
| CN115878882A (en) | Hierarchical representation learning of user interests | |
| CN113590673A (en) | Data heat degree statistical method based on block chain deep learning | |
| CN111723301A (en) | Attention relation identification and labeling method based on hierarchical theme preference semantic matrix | |
| CN113935804B (en) | Semantic recommendation method based on reinforcement learning and weighting element path | |
| Cuevas et al. | An improved evolutionary algorithm for reducing the number of function evaluations |
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 |