LU503730B1 - Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns - Google Patents

Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns Download PDF

Info

Publication number
LU503730B1
LU503730B1 LU503730A LU503730A LU503730B1 LU 503730 B1 LU503730 B1 LU 503730B1 LU 503730 A LU503730 A LU 503730A LU 503730 A LU503730 A LU 503730A LU 503730 B1 LU503730 B1 LU 503730B1
Authority
LU
Luxembourg
Prior art keywords
sequence
item
vector
relationship
attribute
Prior art date
Application number
LU503730A
Other languages
French (fr)
Inventor
Quangui Zhang
Original Assignee
Univ Chongqing Arts & Sciences
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Univ Chongqing Arts & Sciences filed Critical Univ Chongqing Arts & Sciences
Application granted granted Critical
Publication of LU503730B1 publication Critical patent/LU503730B1/en

Links

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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2457—Query processing with adaptation to user needs
    • G06F16/24578—Query processing with adaptation to user needs using ranking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
    • 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
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • 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
    • G06N3/09—Supervised learning
    • 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
    • G06Q30/01—Customer relationship services
    • G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
    • 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
    • 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
    • G06Q30/0282—Rating or review of business operators or products
    • 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
    • G06N3/048—Activation functions

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Computational Linguistics (AREA)
  • Strategic Management (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • General Business, Economics & Management (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Fuzzy Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Game Theory and Decision Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

A sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns relates to the field of intelligent recommendation technology and fully utilizes the coupling relationship between user and item interactive session sequence mode and item attribute information to effectively improve recommendation accuracy. The technical scheme: build data sets; divide data sets; build sequence recommendation model: learn the interaction time sequence relationship between user and item, the time sequence relationship of item attributes through GRU network, and learn the coupling relationship between item attributes and time sequence mode through attention mechanism network, and then build a sequence recommendation model with candidate item interaction prediction layer; train sequence recommendation model: input training set into the sequence recommendation model which learns the interactive sequence mode combining the coupling relationship between item attributes and sequence mode, and the learning parameters are obtained; sequence recommendation model prediction.

Description

Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns HUS08750
BACKGROUND Technical Field
This invention relates to the field of intelligent recommendation technology, in particular to a sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns.
Related Art
With the widespread use of recommendation systems, the sequence-based recommendation has gradually become an important sub-field in the field of recommendation systems. Unlike traditional recommendation methods, which are mostly based on modeling user information, sequence recommendation methods do not focus on user information, but generate recommendations for users by mining the relationship between items in session sequence data. However, the existing sequence recommendation methods tend to focus too much on the sequence data itself, and ignore the importance of some auxiliary information. They do not make full use of the influence of the item attribute information and the coupling relationship between the item attribute information and the time series on the recommendation results, so the recommendation results are often inaccurate. In fact, there is a coupling relationship between the interactive conversation sequence mode of users and items and the item attribute information. So how to make full use of the coupling relationship between the interactive conversation sequence mode of users and items and the item attribute information to effectively improve the accuracy of recommendations and thus improve user satisfaction is a technical problem that needs to be solved at present.
Content of invention
The technical task of the invention is to provide a sequence recommendation method and system based on the coupling relationship between the item attributes and the time sequence patterns, in order to solve the problem of how to make full use of the coupling relationship between the user and the interactive session sequence pattern of the item and the item attribute information to effectively improve the accuracy of the recommendation, and thus improve the user's satisfaction.
The technical task of the invention is realized by a sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns, which is as follows:
Build data set: Shuffle the interaction history data between users and items, and build the user behavior sequence data set according to the sequence of interaction time between users and items; HUS03730
Divide data set: Divide the user behavior sequence data set into the training set, test set, and verification set; The verification set is used to adjust the hyperparameter of the sequence recommendation model in the training stage of training.
Build a sequence recommendation model: Learn the interaction time sequence relationship between users and items and the time sequence relationship of item attributes through the GRU network, and the coupling relationship between item attributes and time sequence patterns through the attention mechanism network, and then build a sequence recommendation model combining with the interaction prediction layer of candidate items;
Train sequence recommendation model: Input the training set into the sequence recommendation model, and the sequence recommendation model learns the interactive sequence pattern representation of the coupling relationship between the item attributes and the sequence pattern, and then trains the learnable parameters of the sequence recommendation model;
Predict sequence recommendation model: Input the test set into the sequence recommendation model. The sequence recommendation model predicts and sorts the interaction probability of all candidate items, and selects the first K items as the recommendation list.
As selected items, each user behavior sequence data set includes multiple subsequence data sets, which include the item ID subsequence data set and item attribute subsequence data set, and the item attribute subsequence data set is constructed according to each attribute of the item.
Optimally, the sequence recommendation model is built as follows:
Obtain the interaction time sequence relationship between users and items: Use the item ID to build the sequence data of the items that users have interacted with at the time of t, and learn the implicit interaction relationship between users and items which is expressed as vector q through GRU network;
Obtain the temporal relationship of item attributes: the items in the interaction sequence between users and items are formed into multiple attribute sequences according to the attribute of each item, and the temporal relationship expression vector of each item attribute sequence is learned by GRU network;
Obtain the coupling relationship between the item attribute and the time sequence pattern: take the implicit interaction expression vector q of the user and the item as the query vector, and use the attention mechanism network to learn the time sequence relationship of each item attribute sequence to express the coupling relationship between 909750 the vector and the query vector, so as to generate the final sequence representation vector based on the coupling relationship analysis;
Interaction prediction of candidate items: After the similarity calculation between the final sequence expression vector based on coupling relationship analysis and the embedded vector of candidate items, the final interaction probability of each candidate item and the user is input in the fully connected network, and the Top-K recommendation is generated according to the ranking result of interaction probability.
Optimally, obtaining the time sequence relationship of item attributes is as follows:
Build item attribute sequence: According to the interaction time sequence between users and items, each attribute value of items in the sequence data is formed into an attribute sequence, such as item brand sequence, item category sequence, etc;
The attribute values in each attribute sequence are represented by one-hot coding, and are converted into low-dimensional dense vectors, namely embedded vectors, respectively through single-layer full-connected networks;
Input each attribute sequence into the GRU network and learn the representation vector sai of each attribute sequence.
Optionally, the method of obtaining the coupling relationship between the item attributes and the time pattern is as follows:
The implicit interaction relationship expression vector q between users and items is used as the query vector, and the implicit interaction relationship expression vector q between users and items and the expression vector sai of each attribute sequence is calculated by similarity through dot product to obtain the weight, and the normalized weight aiis calculated using Softmax function, and the formula is as follows: a; = softmax(dot(Wdq, Wksai)) :
Weight ai and the corresponding attribute sequence vector are weighted and summed to obtain the vector s, s = Ni (WYSai) : wherein WI, UK, and W are transformation matrices, which are learnable parameters;
Vector s further learns through one or multi-layers of the fully-connected network to get the joint relationship c of final product attributes and time sequence pattern.
The interaction prediction between the candidates is as follows:
Calculation of resemblance: after changing the ID of the candidate item i through the single-layer fully-connection network into a low-dimension dense vector, this vector and the final relationship between the item attributes and the time sequence pattern ¢ would be multiplied to calculate the similarity vector d, the relationship between the attributes ? 99799 and time sequence patterns of candidate items and final items;
Similarity vector d is further learned through one or multi-layers of fully connected networks;
The Sigmoidal activation function is used to compress the output to the [0,1] range to get the probability that the candidate will be an object for user interaction at t+1, that is, the final prediction result.
A sequence recommendation system based on the coupling relationship between item attributes and timing patterns includes: the data set building unit, which is used to shuffle the interaction history data between users and items and build the user behavior sequence data set according to the sequence of interaction time between users and items; the data set division unit, which is used to divide the user behavior sequence data set into the training set, test set, and verification set; the model building unit, which is used to learn the interaction sequence relationship between users and items and the temporal relationship of item attributes through the GRU network, and the coupling relationship between item attributes and time patterns through the attention mechanism network, and then build a sequence recommendation model combining with the interaction prediction layer of candidate items; the model training unit, which is used to input the training set into the sequence recommendation model which learns the interactive sequence pattern of the coupling relationship between the item attributes and the sequence pattern, further training the learnable parameters of the sequence recommendation model; the prediction unit, which is used to input the test set into the sequence recommendation model that predicts and sorts the interaction probability of all candidate items, and selects the first K items as the recommendation list.
As selected items, the model building unit comprises, the user and item interaction timing relationship acquisition module, which is used to construct sequence data from the item ID that the user has interacted with at t times, and uses the GRU network to learn the vector q of implicit interaction relationship between the user and the item; the item attribute temporal relationship acquisition module, which is used to form multiple attribute sequences according to each attribute of the item in the interaction sequence between the user and the item, and uses the GRU network to learn the temporal relationship expression vector of each item attribute sequence; the coupling relationship acquisition module of item attributes and temporal patterns 599750 which is used to take the implicit interaction relationship expression vector q of users and items as the query vector, and uses the attention mechanism network to learn the coupling relationship between the temporal relationship expression vector of each item attribute sequence and the query vector, So as to generate the final sequence vector based on the coupling relationship analysis; the candidate item interaction prediction module, which is used to calculate the similarity between the final sequence vector based on coupling analysis and the candidate item embedding vector, and then inputs it into the fully-connected network to finally generate the interaction probability of each candidate item and the user, and generates Top-K recommendations according to the interaction probability ranking results.
Optimally, the acquisition module of the temporal relationship of the article attributes includes, the sub-module of item attribute sequence construction, which is used to form the attribute sequence of each attribute value of the item in the sequence data according to the interaction time sequence between the user and the item, such as item brand sequence, item category sequence, etc; encoding and conversion sub-module, which is used to express the attribute values in each attribute sequence using one-hop coding, and converts them into low-dimensional dense vectors, namely, embedded vectors, through a single-layer full-connected network; the learning sub-module 1, which is used to input each attribute sequence into the
GRU network and learn the vector sai of each attribute sequence; the acquisition module of the coupling relationship between the item attribute and the time sequence pattern includes, the dot product sub-module, which is used to take the implicit interaction vector q of the user and the item as the query vector. Vector q and the vector sai of each attribute sequence calculate the weight through the similarity calculation of the dot product, and use the Softmax function to calculate the normalized weight ai. The formula is as follows: a; = softmax(dot(Wdq, WKs,;));
Weighted summation sub-module, weight ai, and the corresponding attribute sequence vector are weighted and summed to obtain the vector s, s = Zi (WYsai) 3 wherein WI, UK, and W are transformation matrices, which are learnable parameters; the learning sub-module 2 is used to further learn vector s through one or multi-layer fully- connected network to get the final product attributes and time series pattern collusion relationship c. 7908780
The described prediction units include:
Calculation of resemblance, after changing the ID of the candidate item i through the single-layer fully-connection network into a low-dimension dense vector, this vector and the final relationship between the item attributes and the time sequence pattern c would be multiplied to calculate the similarity vector d, the relationship between the attributes and time sequence patterns of candidate items and final items; the learning sub-module 3, used to further learn the similarity vector d through one or multi-layers of the fully-connected network; compression sub-module, used to compress the output to the [0,1] range using the
Sigmoid activation function to get the probability that candidates will be the item for user interaction at t+1 time, that is, the final prediction result.
An electronic device includes memory that stores computer programs and at least one processor;
At least one processor executes a computer program stored in the memory such that at least one processor executes the sequence recommendation method based on the coupling relationship between item attributes and time series mode as described above.
A computer-readable storage medium in which a computer program is stored can be executed by a processor to implement the sequence recommendation based on the coupling of item attributes and time sequence modes as described above.
The invention of a sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns has the following advantages: (1) This invention incorporates the interactive sequence mode between users and items and the coupling relationship between each attribute of the items into the sequence recommendation model, and makes full use of the coupling relationship between the interactive session sequence mode of the user and the item and the attribute information of the item to effectively improve the accuracy of the recommendation, thereby improving the user's satisfaction. (2) This invention achieves Top-K recommendation by constructing and training a learning network consisting of item time sequence relationship, item attribute time series relationship and coupling relationship, and a network model consisting of an interactive prediction layer for candidate objects to learn the interactive sequence mode that fuses the coupling relationship between item attributes and time sequence mode to generate a prediction of the interaction probability of candidate items. (3) This invention considers the coupling relationship between the user and the term 0 interactive sequence mode and the item attributes, and improves the accuracy of the recommendation. (4) This invention considers the influence of the subsequence formed by each property of the item on the final recommendation result respectively. (5) This invention effectively combines the cyclic neural network, attention mechanism, and collaborative filtering to improve the interpretability of sequence recommendation.
Figure Description
This invention is further described below with the figures.
Fig. 1 is a flow diagram of a sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns.
Fig. 2 is a flow chart for building a sequence recommendation model.
Fig. 3 is a block diagram of a sequence recommendation system based on the coupling relationship between item attributes and time sequence patterns.
Fig. 4 illustrates a sequence recommendation model.
Specific Example methods
The sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns will be described in detail below with reference to the drawings and specific examples.
Example 1:
As shown in Fig. 1, the present example provides a sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns, which is as follows:
S1. Build data set: shuffle the interaction history data between users and items, and build the user behavior sequence data set according to the sequence of interaction time between users and items; each user behavior sequence data set includes multiple subsequence data sets that include the item ID subsequence data set and the item attribute subsequence data set, and the item attribute subsequence data set is constructed according to each item attribute.
Take the Amazon data set as an example to show the user-item interaction history
LU503730 data:
Field Field meaning Example reviewerlD user ID A2SUAM1J3GNN3B asin product ID 0000013714 reviewerName user name J. McDonald helpful effective evaluation [2, 3] rate reviewText comment text Great purchase though! overall grade 5.0 summary evaluation summary Heavenly Hignway Hymns unixReviewTime evaluation timestamp 1252800000 reviewTime evaluation time 09 13, 2009
An example of item attribute information is shown as follows:
Field Field meaning Example asin product ID 0000031852 title product name Girls Ballet Tutu Zebra Hot Pink price price 3.17 http://ecx.images- imUrl product image link amazon.com/images/l/51fAmVkTbyL._S
Y300_.jpg {"also_bought": ["BOOJHONN1S", "B002BZX826"], related relevant products "also_viewed". [ BO02BZX876", pP "BOOJHONN1S"], "bought_together": ["B0O02BZX8Z6"1} salesRank discount information {"Toys & Games": 211836} brand brand Coxlures cateaories cateao [['Sports & Outdoors", "Other 9 gory Sports", "Dance']]
S2. Divide the data set: divide the user behavior sequence data set into the training set, test set, and verification set according to the ratio of 6:2:2;
S3. Build a sequence recommendation model: learn the interaction time sequence relationship between users and items and the time sequence relationship of item attributes through the GRU network, and the coupling relationship between item attributes and time sequence patterns through the attention mechanism network, and then build a sequence recommendation model combining with the interaction prediction layer of candidate items;
S4. Train sequence recommendation model: input the training set into the sequence recommendation model, and the sequence recommendation model learns the interactive sequence pattern of the coupling relationship between the item attributes and the sequence pattern, further training the learnable parameters of the sequence recommendation model;
S5. Sequence recommendation model prediction: input the test set into the sequence 0 recommendation model which predicts and sorts the interaction probability of all candidate items, and selects the first K items as the recommendation list.
As shown in Fig. 2, the construction sequence recommendation model in S3 of this example is as follows:
S301. Obtain the interaction time sequence relationship between users and items: use the item ID to build sequence data of the items that users have interacted with at the time of t, and use the GRU network to learn the implicit interaction relationship between users and items as vector q;
S302. Obtain the temporal relationship of item attributes: the items in the interaction sequence between users and items are formed into multiple attribute sequences according to each item attribute, and the temporal relationship vector of each item attribute sequence is learned using the GRU network;
S303. Obtain the coupling relationship between item attributes and time sequence mode: use the implicit interaction between user and item vector Q as the query vector, and use the attention mechanism network to learn the coupling relationship between the time sequence vector and query vector of each item attribute sequence, and generate the final sequence vector based on coupling relationship analysis;
S304. Interactive prediction of candidate items: After similarity calculation between the final sequence vector based on coupling analysis and the embedded vector of the candidate items, input the interaction probabilities between each candidate and user generated in the fully-connected network, and Top-K recommendations are generated based on the sorting results of the interaction probabilities.
The method for obtaining the time sequence relationship of the item attributes in
S302, Example 2, is as follows:
S30201. Construct the attribute sequence of item: According to the time sequence relationship between the user and the item, each attribute value of an item in the serial data forms an attribute sequence, such as item brand sequence, category sequence, etc;
S30202. The attribute values in each attribute sequence are represented by one-hot encoding and converted into low-dimensional dense vectors, i.e., embedded vectors, through a single layer fully-connected network;
S30203. Input each attribute sequence into the GRU network and learn the vector sai of each attribute sequence.
In S303 of this example, the coupling relationship between getting item attributes and time sequence patterns is as follows:
S30301. Take the implicit interaction vector q of the user and the item as the query 999790 vector. Vector q and the vector sai of each attribute sequence calculate the weight through the similarity calculation of the dot product, and use the Softmax function to calculate the normalized weight ai. The formula is as follows: a; = softmax(dot(Wdq, Wks,;)) :
S30302. Weight ai and the corresponding attribute sequence vector are weighted and summed to obtain the vector s, s = Xi aq (WYsai) 3 wherein WI, UK, and W are transformation matrices, which are learnable parameters;
S30303. The vector s is further learned through one or multi-layers of the fully- connected network to obtain the final product attribute and time sequence pattern summation relationship c;
The interaction prediction of candidate items in S304 of this example is as follows:
S30401. Calculation of resemblance: after changing the ID of the candidate item i through the single-layer fully-connection network into a low-dimension dense vector, this vector and the final relationship between the item attributes and the time sequence pattern c would be multiplied to calculate the similarity vector d, the relationship between the attributes and time sequence patterns of candidate items and final items;
S30402. Similarity vector d is further learned through one or multi-layers of fully connected networks:
S30403. The Sigmoidal activation function is used to compress the output to the [0,1] range to get the probability that the candidate will be an object for user interaction at t+1, that is, the final prediction result.
Example 2:
As shown in Fig.3, this example provides a sequence recommendation system based on the coupling relationship between item attributes and time sequence patterns, which includes: the data set building unit, which is used to shuffle the interaction history data between users and items and build the user behavior sequence data set according to the sequence of interaction time between users and items; the data set division unit, which is used to divide the user behavior sequence data set into training set, test set and verification set; the model building unit, which is used to learn the interaction sequence relationship between users and items and the temporal relationship of item attributes through the GRU network, and the coupling relationship between item attributes and time patterns through 997/90 the attention mechanism network, and then build a sequence recommendation model combining with the interaction prediction layer of candidate items; the model training unit, which is used to input the training set into the sequence recommendation model which learns the interactive sequence pattern of the coupling relationship between the item attributes and the sequence pattern, further training the learnable parameters of the sequence recommendation model; the prediction unit, which is used to input the test set into the sequence recommendation model that predicts and sorts the interaction probability of all candidate items, and selects the first K items as the recommendation list.
The model building unit in this example includes: the user and item interaction timing relationship acquisition module, which is used to construct sequence data from the item ID that the user has interacted with at t times, and uses the GRU network to learn the vector q of implicit interaction relationship between the user and the item; the item attribute temporal relationship acquisition module, which is used to form multiple attribute sequences according to each attribute of the item in the interaction sequence between the user and the item, and uses the GRU network to learn the temporal relationship expression vector of each item attribute sequence; the coupling relationship acquisition module of item attributes and temporal patterns, which is used to take the implicit interaction relationship expression vector q of users and items as the query vector, and uses the attention mechanism network to learn the coupling relationship between the temporal relationship expression vector of each item attribute sequence and the query vector, so as to generate the final sequence vector based on the coupling relationship analysis; the candidate item interaction prediction module, which is used to calculate the similarity between the final sequence vector based on coupling analysis and the candidate item embedding vector, and then inputs it into the fully-connected network to finally generate the interaction probability of each candidate item and the user, and generate Top-K recommendations according to the interaction probability ranking results.
As shown in Fig. 4, the item attribute timing relationship acquisition modules in this example include: the sub-module of item attribute sequence construction, which is used to form the attribute sequence of each attribute value of the item in the sequence data according to the interaction time sequence between the user and the item, such as item brand sequence, item category sequence, etc; HUS08750 encoding and conversion sub-module, which is used to express the attribute values in each attribute sequence using one-hop coding, and converts them into low-dimensional dense vectors, namely, embedded vectors, through the single-layer full-connected network; the learning sub-module 1, which is used to input each attribute sequence into the
GRU network and learn the vector sai of each attribute sequence. the acquisition module of the coupling relationship between the item attribute and the time sequence pattern in this example includes, the dot product sub-module, which is used to take the implicit interaction vector q of the user and the item as the query vector. Vector q and the vector sai of each attribute sequence calculate the weight through the similarity calculation of the dot product, and use the Softmax function to calculate the normalized weight ai. The formula is as follows: a; = softmax(dot(Wdq, Wks,;)) :
Weighted summation sub-module, weight ai, and the corresponding attribute sequence vector are weighted and summed to obtain the vector s, s = Zi (WYsai) 3 wherein WI, UK, and W are transformation matrices, which are learnable parameters; the learning sub-module 2 is used to further learn vector s through one or multi-layer fully- connected network to get the final product attributes and time series pattern collusion relationship c.
The prediction units in this example include:
Similarity calculation sub-module, after changing the ID of the candidate item i through the single layer fully-connection network into a low dimension dense vector, this vector and the final relationship between the item attributes and the time sequence pattern c would be multiplied to calculate the similarity vector d, the relationship between the attributes and time sequence patterns of candidate items and final items;
The learning sub-module 3, used to further learn the similarity vector d through one or multi-layers of the fully-connected network;
Compression sub-module, used to compress the output to the [0,1] range using the
Sigmoid activation function to get the probability that candidates will be the item for user interaction at t+1 time, that is, the final prediction result.
Example 3:
This example provides an electronic device including a memory that stores computer programs and at least one processor;
The processor executes the computer execution instructions stored in the memory. 0 making the processor execute the sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns in any example of the invention.
The processor can be a central processing unit (CPU), other general purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor can be a microprocessor or any conventional processor.
The memory can be used to store computer programs and/or modules. The processor can realize various functions of electronic equipment by running or executing computer programs and/or modules stored in the memory and calling data stored in the memory. The memory can mainly include the storage program area and the storage data area, wherein the storage program area can store the operating system, the application program required by at least one function, etc; The storage data area can store the data created according to the use of the terminal. In addition, the memory can also include high-speed random access memory and non-volatile memory, such as hard disk, memory, plug-in hard disk, Smart Media Card (SMC), secure digital (SD) card, flash memory card, at least one disk storage period, flash memory device, or other volatile solid-state storage devices.
Example 4:
This example also provides a computer-readable storage medium, in which a plurality of instructions are stored, and the instructions are loaded by the processor to enable the processor to execute the sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns in any example of the invention. Specifically, a system or device equipped with a storage medium can be provided, and the storage medium is stored with software program code to realize the functions of any of the above examples, and the computer (or CPU or MPU) of the system or device can read and execute the program code stored in the storage medium.
In this case, the program code read from the storage medium itself can realize the functions of any of the above examples, so the program code and the storage medium storing the program code form a part of the present invention.
Examples of storage media used to provide program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM,
DVD-RYM, DVD-RW, and DVD+RW), magnetic tapes, non-volatile memory cards, and
ROMs. Optionally, the program code can be downloaded from the server computer by the 503790 communication network.
In addition, it should be clear that the operating system operating on the computer completes some or all of the actual operations not only through executing the program code read by the computer, but also through using the instructions based on the program code, thus realizing the functions of any of the above examples.
Besides, it can be understood that the program code read out by the storage medium is written to the memory set in the expansion board inserted in the computer or the memory set in the expansion unit connected to the computer, and then based on the instructions of the program code, the CPU installed on the expansion board or the expansion unit performs part and all of the actual operations, thus realizing the functions of any of the above examples.
Finally, it should be noted that the above examples are only used to illustrate the technical solution of the invention, not to limit it; Although the present invention has been described in detail with reference to the preceding examples, ordinary technical personnel in the art should understand that they can still modify the technical solutions recorded in the preceding examples, or replace some or all of the technical features equally; However, these modifications or replacements do not make the essence of the corresponding technical solutions separate from the scope of the technical solutions of the examples of the invention.

Claims (10)

CLAIMS LU503730
1. A sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns, wherein: Build data set: Shuffle the interaction history data between users and items, and build the user behavior sequence data set according to the sequence of interaction time between users and items; Divide data set: Divide the user behavior sequence data set into the training set, test set, and verification set; Build a sequence recommendation model: Learn the interaction time sequence relationship between users and items and the time sequence relationship of item attributes through the GRU network, and the coupling relationship between item attributes and time sequence patterns through the attention mechanism network, and then build a sequence recommendation model combining with the interaction prediction layer of candidate items; Train sequence recommendation model: Input the training set into the sequence recommendation model, and the sequence recommendation model learns the interactive sequence pattern representation of the coupling relationship between the item attributes and the sequence pattern, and then trains the learnable parameters of the sequence recommendation model; Predict sequence recommendation model: Input the test set into the sequence recommendation model. The sequence recommendation model predicts and sorts the interaction probability of all candidate items, and selects the first K items as the recommendation list.
2. A sequence recommendation method based on the coupling relationship between item attributes and time sequence patterns according to claim 1, characterized in that each user behavior sequence data set includes multiple subsequence data sets, which include item ID subsequence data set and item attribute subsequence data set, and the item attribute subsequence data set is constructed according to each attribute of the item.
3. A sequence recommendation method based on the coupling relationship between item attributes and time sequence model according to claim 1 and claim 2, characterized in that a sequence recommendation model is built as follows: Obtain the interaction time sequence relationship between users and items: Use the item ID to build the sequence data of the items that users have interacted with at the time of t, and learn the implicit interaction relationship between users and items which is expressed as vector q through GRU network; Obtain the temporal relationship of item attributes: the items in the interaction 99/99 sequence between users and items are formed into multiple attribute sequences according to the attribute of each item, and the temporal relationship vector of each item attribute sequence is learned by GRU network; Obtain the coupling relationship between the item attribute and the time sequence patterns: take the implicit interaction vector q of the user and the item as the query vector, and use the attention mechanism network to learn the time sequence relationship of each item attribute sequence to express the coupling relationship between the vector and the query vector, So as to generate the final sequence representation vector based on the coupling relationship analysis; Interaction prediction of candidate items: After the similarity calculation between the final sequence expression vector based on coupling relationship analysis and the embedded vector of candidate items, the final interaction probability of each candidate item and the user is input in the fully-connected network, and the Top-K recommendation is generated according to the ranking result of interaction probability.
4. A sequence recommendation method based on the coupling relationship between the item attributes and the time sequence patterns according to claim 3, characterized in that the time sequence relationship for obtaining the item attributes is as follows: Build item attribute sequence: According to the interaction time sequence between users and items, each attribute value of items in the sequence data is formed into an attribute sequence, such as item brand sequence, item category sequence, etc; The attribute values in each attribute sequence are represented by one-hot coding, and are converted into low-dimensional dense vectors, namely embedded vectors, respectively through single-layer full-connected networks; Input each attribute sequence into the GRU network and learn the representation vector sai of each attribute sequence.
5. According to the sequence recommendation method based on the coupling relationship between the item attribute and the time sequence patterns according to claim 4, characterized in that the coupling relationship between the item attribute and the time sequence model is obtained as follows: The implicit interaction relationship expression vector q between users and items is used as the query vector, and the implicit interaction relationship expression vector q between users and items and the expression vector sai of each attribute sequence are calculated by similarity through dot product to obtain the weight, and the normalized weight aiis calculated using Softmax function, and the formula is as follows: a; = softmax(dot(Wdq, Wksai)) : 7593799 Weight ai and the corresponding attribute sequence vector are weighted and summed to obtain the vector s, s = Xi aq (WYsai) 3 wherein WI, UK, and W are transformation matrices, which are learnable parameters; Vector s further learns through single or multi-layers of the fully-connected network to get the joint relationship c of final product attributes and time sequence pattern; The interaction prediction of candidate items is as follows: Calculation of resemblance: after changing the ID of the candidate item i through the single-layer fully-connected network into a low-dimension dense vector, this vector and the final relationship between the item attributes and the time sequence pattern c would be multiplied to calculate the similarity vector d, the relationship between the attributes and time sequence patterns of candidate items and final items; The similarity vector d is further learned through one or multi-layer fully-connected networks: The Sigmoidal activation function is used to compress the output to the [0,1] range to get the probability that the candidate will be an object for user interaction at t+1, that is, the final prediction result.
6. A sequence recommendation system based on the coupling relationship between item attributes and time sequence patterns, wherein, the system includes: the data set building unit, which is used to shuffle the interaction history data between users and items and build the user behavior sequence data set according to the sequence of interaction time between users and items; the data set division unit, which is used to divide the user behavior sequence data set into the training set, test set, and verification set; the model building unit, which is used to learn the interaction sequence relationship between users and items and the temporal relationship of item attributes through the GRU network, and the coupling relationship between item attributes and time patterns through the attention mechanism network, and then build a sequence recommendation model combining with the interaction prediction layer of candidate items; the model training unit, which is used to input the training set into the sequence recommendation model which learns the interactive sequence pattern of the coupling relationship between the item attributes and the sequence pattern, further training the learnable parameters of the sequence recommendation model;
the prediction unit, which is used to input the test set into the sequence recommendation model that predicts and sorts the interaction probability of all candidate 99/50 items, and selects the first K items as the recommendation list.
7. A sequence recommendation system based on the coupling relationship between item attributes and time sequence patterns according to claim 6, wherein, the model building units include: the user and item interaction timing relationship acquisition module, which is used to construct sequence data from the item ID that the user has interacted with at t times, and uses the GRU network to learn the vector q of implicit interaction relationship between the user and the item; the item attribute temporal relationship acquisition module, which is used to form multiple attribute sequences according to each attribute of the item in the interaction sequence between the user and the item, and uses the GRU network to learn the temporal relationship expression vector of each item attribute sequence; the coupling relationship acquisition module of item attributes and temporal patterns, which is used to take the implicit interaction relationship expression vector q of users and items as the query vector, and uses the attention mechanism network to learn the coupling relationship between the temporal relationship expression vector of each item attribute sequence and the query vector, so as to generate the final sequence vector based on the coupling relationship analysis; the candidate item interaction prediction module, which is used to calculate the similarity between the final sequence vector based on coupling analysis and the candidate item embedding vector, and then inputs it into the fully-connected network to finally generate the interaction probability of each candidate item and the user, and generates Top-K recommendations according to the interaction probability ranking results.
8. A sequence recommendation system based on the coupling relationship between the item attributes and the time sequence pattern according to claim 7, characterized in that the item attribute time relationship acquisition patterns include: the sub-module of item attribute sequence construction, which is used to form the attribute sequence of each attribute value of the item in the sequence data according to the interaction time sequence between the user and the item; encoding and conversion sub-module, which is used to express the attribute values in each attribute sequence using one-hop coding, and converts them into low-dimensional dense vectors, namely, embedded vectors, through the single-layer full-connected network;
the learning sub-module 1, which is used to input each attribute sequence into the GRU network and learn the vector sai of each attribute sequence; HUS08750 The acquisition module of the coupling relationship between the item attribute and the time sequence pattern includes, the dot product sub-module, which is used to take the implicit interaction vector q of the user and the item as the query vector. Vector q and the vector sai of each attribute sequence calculate the weight through the similarity calculation of the dot product, and use the Softmax function to calculate the normalized weight ai. The formula is as follows: a; = softmax(dot(Wdq, Wksai)) : Weighted summation sub-module, weight ai, and the corresponding attribute sequence vector are weighted and summed to obtain the vector s, s = Zi (WYsai) 3 wherein WA, AK, and W are transformation matrices, which are learnable parameters; the learning sub-module 2 is used to further learn vector s through one or multi-layer fully-connected network to get the final product attributes and time series pattern collusion relationship c. The prediction unit comprises: similarity calculation sub-module, after changing the ID of the candidate item i through the single layer fully-connection network into a low-dimension dense vector, this vector and the final relationship between the item attributes and the time sequence pattern c would be multiplied to calculate the similarity vector d, the relationship between the attributes and time sequence patterns of candidate items and final items; learning sub-module 3, used to further learn the similarity vector d through one or multi-layers of the fully-connected network; compression sub-module, used to compress the output to the [0,1] range using the Sigmoid activation function to get the probability that candidates will be the item for user interaction at t+1 time, that is, the final prediction result.
9. An electronic device characterized in that it includes a memory that stores computer programs and at least one processor; At least one processor executes a computer program stored in the memory such that at least one processor executes the sequence recommendation method based on the coupling relationship between item attributes and time sequence pattern according to claims 1-5.
10. A computer-readable storage medium characterized in that a computer program is stored in it and can be executed by a processor to implement the sequence recommendation based on the coupling of item attributes and time sequence patterns LU503730 according to claims 1-5.
LU503730A 2022-09-26 2023-01-05 Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns LU503730B1 (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211171763.3A CN115601098A (en) 2022-09-26 2022-09-26 Sequence recommendation method and system based on coupling relation between article attributes and time sequence modes

Publications (1)

Publication Number Publication Date
LU503730B1 true LU503730B1 (en) 2023-10-27

Family

ID=84844082

Family Applications (1)

Application Number Title Priority Date Filing Date
LU503730A LU503730B1 (en) 2022-09-26 2023-01-05 Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns

Country Status (3)

Country Link
CN (1) CN115601098A (en)
LU (1) LU503730B1 (en)
WO (1) WO2023138381A1 (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116703529B (en) * 2023-08-02 2023-10-20 山东省人工智能研究院 Contrast learning recommendation method based on feature space semantic enhancement

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200311585A1 (en) * 2019-03-31 2020-10-01 Palo Alto Networks Multi-model based account/product sequence recommender
CN112084407B (en) * 2020-09-08 2024-03-12 辽宁工程技术大学 Collaborative filtering recommendation method integrating graph neural network and attention mechanism
CN112269927B (en) * 2020-10-22 2024-05-28 辽宁工程技术大学 Recommendation method based on session sequence dynamic behavior preference coupling relation analysis
CN112328893B (en) * 2020-11-25 2022-08-02 重庆理工大学 Recommendation method based on memory network and cooperative attention

Also Published As

Publication number Publication date
WO2023138381A1 (en) 2023-07-27
CN115601098A (en) 2023-01-13

Similar Documents

Publication Publication Date Title
Liu et al. Physical-virtual collaboration modeling for intra-and inter-station metro ridership prediction
Miech et al. Learnable pooling with context gating for video classification
WO2021139415A1 (en) Data processing method and apparatus, computer readable storage medium, and electronic device
CN110364146A (en) Audio recognition method, device, speech recognition apparatus and storage medium
CN109783539A (en) Usage mining and its model building method, device and computer equipment
CN112380453B (en) Article recommendation method and device, storage medium and equipment
CN112307351A (en) Model training and recommending method, device and equipment for user behavior
LU503730B1 (en) Sequence recommendation method and system based on the coupling relationship between item attributes and time sequence patterns
CN112699310A (en) Cold start cross-domain hybrid recommendation method and system based on deep neural network
CN113761250A (en) Model training method, merchant classification method and device
CN112527993A (en) Cross-media hierarchical deep video question-answer reasoning framework
CN112269943B (en) Information recommendation system and method
CN116861258B (en) Model processing method, device, equipment and storage medium
CN116935170B (en) Processing method and device of video processing model, computer equipment and storage medium
CN113032676A (en) Recommendation method and system based on micro-feedback
CN116980665A (en) Video processing method, device, computer equipment, medium and product
Li et al. Directional attention weaving for text-grounded conversational question answering
He et al. Interest HD: An interest frame model for recommendation based on HD image generation
CN115329183A (en) Data processing method, device, storage medium and equipment
CN114996435A (en) Information recommendation method, device, equipment and storage medium based on artificial intelligence
CN113626564B (en) Concept label generation method and device, electronic equipment and storage medium
CN114218405B (en) Knowledge extraction method, related device, electronic equipment and medium
CN117556150B (en) Multi-target prediction method, device, equipment and storage medium
TOUHAMI et al. Session-based Recommendation Systems with Graph ATtention Networks
CN114936563B (en) Event extraction method, device and storage medium

Legal Events

Date Code Title Description
FG Patent granted

Effective date: 20231027