CN110807156A - Interest recommendation method and system based on user sequence click behaviors - Google Patents
Interest recommendation method and system based on user sequence click behaviors Download PDFInfo
- Publication number
- CN110807156A CN110807156A CN201911012817.XA CN201911012817A CN110807156A CN 110807156 A CN110807156 A CN 110807156A CN 201911012817 A CN201911012817 A CN 201911012817A CN 110807156 A CN110807156 A CN 110807156A
- Authority
- CN
- China
- Prior art keywords
- session
- interest
- user
- sequence
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 230000006399 behavior Effects 0.000 title claims abstract description 36
- 230000002452 interceptive effect Effects 0.000 claims abstract description 39
- 230000008846 dynamic interplay Effects 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 15
- 230000003993 interaction Effects 0.000 claims description 34
- 238000000605 extraction Methods 0.000 claims description 17
- 230000004913 activation Effects 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims 1
- 239000011159 matrix material Substances 0.000 description 10
- 230000007246 mechanism Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000001914 filtration Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000007774 longterm Effects 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 241000288105 Grus Species 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses an interest recommendation method and system based on user sequence clicking behaviors, which comprises the steps of firstly, obtaining historical interactive project data of a user to form a historical interactive project sequence of the user; constructing an interest recommendation model; carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model; extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user; interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations; and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
Description
Technical Field
The invention relates to the technical field of project recommendation, in particular to an interest recommendation method and system based on user sequence click behaviors.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
With the rapid development of the internet industry, we have entered an information explosion era. The acceptance of individuals is severely "overloaded" by a wide variety of items, rapidly increasing news information, advertising information spread over the sky, and the like. The enormous amount of information on the internet brings great challenges to both information providers and information users: how an information provider aims at showing mass information stored by the information provider to an information user; how a user of information screens out information needed by himself from a plurality of information. The recommendation system is generated according to the series of problems.
So-called recommendation systems, in short, recommend content for a user that is most likely to be of interest based on the user's preferences. The general recommendation method is mainly represented by a collaborative filtering model. The most common collaborative filtering methods mainly include a neighborhood method and a matrix decomposition model. The neighborhood method carries out recommendation by measuring the distance between the items and the similarity matrix, and the algorithms based on the neighborhoods are divided into two categories, one category is a collaborative filtering algorithm based on users, and the other category is a collaborative filtering algorithm based on articles. While matrix factorization estimates user preferences for items by the inner product between user and item embedding. Although the two methods described above have achieved significant results in terms of recommendation, these methods simply refer to items that the user has interacted with as a collection of items, and recommend them by analyzing the collection. In short, they simply recommend by mining static correlations between users and items. They only focus on the long-term static preferences of the user, ignore the transition of preferences in the hidden user sequence behavior, and are not able to model complex relationships in the sequence data, and are therefore not suitable for sequence recommendation. The sequential recommendation does not simply recommend items interacted with by the user as a collection of items, but rather recommends items as a sequence of items, taking into account not only the long-term static general preferences of the user, but also the short-term dynamic sequential patterns of the user.
In order to solve the problems of the general recommendation system, the sequence recommendation system based on the user interaction behavior has received extensive attention in recent years in academia and industry. The purpose of the sequential recommendation system is to predict the successive items that a user is likely to interact with in the future given the items that the user has interacted with in the past. When people surf the internet, a great deal of user interaction (such as clicking, commenting, collecting and the like of the user) is recorded, and the rich user interaction behaviors provide opportunities for knowing the interests and hobbies of the user. People effectively recommend the users by analyzing the sequential behaviors of the users and mining the user preferences implicit in the interaction behaviors.
In recent years, as a branch of machine learning, deep learning has become a considerable force, and has started to play a large role in the fields of machine translation, image recognition, voice recognition, and the like. Meanwhile, deep learning models, such as RNN, GRU, LSTM, etc., are also beginning to be used step by step in the field of sequence recommendation. At the same time, some studies have also begun to attempt to improve the performance and interpretability of sequence recommendations using attention-driven mechanisms due to their strong potential to model sequential data in tasks such as machine translation. During the development process, the inventor finds that the general sequence recommendation system takes the item interacted by the user as a whole as an input, firstly, the general sequence recommendation system does not consider the internal structure of the sequence, namely the sequence is composed of sessions, and the interest of the user in the same session is different from one session to another: user behavior is similar in each session and different between different sessions. Second, they only consider the sequence patterns between items and ignore the sequence transformation relationships between items.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an interest recommendation method and system based on user sequence click behaviors, firstly, according to the fact that the preferences of users in a certain time period are similar, the interaction sequences of the users are divided into different conversations, then, the sequence recommendation is realized by combining the user interests in each conversation and the interest interaction among different conversations, and the problems that the existing sequence recommendation method ignores the internal structure of user sequence behaviors and ignores the conversion relation among items are effectively solved.
The technical scheme of the interest recommendation method based on the user sequence click behavior provided by the invention is as follows:
an interest recommendation method based on user sequence click behaviors comprises the following steps:
acquiring historical interactive project data of a user to form a historical interactive project sequence of the user;
constructing an interest recommendation model;
carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model;
extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user;
interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations;
and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
The technical scheme of the interest recommendation system based on the user sequence click behavior provided by the invention is as follows:
an interest recommendation system based on user sequence click behavior, the system comprising:
the data acquisition module is used for acquiring historical interactive project data of the user to form a historical interactive project sequence of the user;
the model construction module is used for constructing an interest recommendation model;
the session division module is used for carrying out session division on the historical interactive item sequence of the user by utilizing the interest recommendation model;
the intra-session interest extraction module is used for extracting each intra-session interest obtained after division;
the activation module is used for endowing different weights to the interests in each session to obtain a session interest sequence of the user;
the inter-session interest interaction module is used for interacting the interests among different sessions to obtain a dynamic interaction model among the different sessions;
and the prediction module is used for inputting the session interest sequence of the user into a dynamic interaction model among different sessions to predict and obtain a target item sequence to be recommended.
Another aspect of the present invention provides a computer-readable storage medium, wherein:
a computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:
acquiring historical interactive project data of a user to form a historical interactive project sequence of the user;
constructing an interest recommendation model;
carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model;
extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user;
interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations;
and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
Another aspect of the present invention provides a processing apparatus, including:
a processing apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps when executing the program:
acquiring historical interactive project data of a user to form a historical interactive project sequence of the user;
constructing an interest recommendation model;
carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model;
extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user;
interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations;
and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
Through the technical scheme, the invention has the beneficial effects that:
(1) the invention divides the click sequence of the user into different conversations, takes the conversation as the minimum operation unit, and can reduce the information loss caused by neglecting or destroying the conversation structure to the maximum extent; and the conversation is used as a basic operation unit, which is more beneficial to parallel processing, and the relative method is very efficient in training efficiency.
(2) After dividing the interaction sequence of the user into different sessions, the embodiment first utilizes an interest extraction layer in a transform session based on multi-head self-attention to extract the interest of the user in each session; compared with the traditional modeling of an interactive sequence through an RNN structure, the relevance of the interest extraction layer in the multi-head self-attention-based Transformer session does not depend on the distance and the distance of the current item and other items no matter how far the spatial distance of the current item and other items is, but depends on the relevance of the two items, so that the sequence item can be better modeled; meanwhile, the attention mechanism is only calculated by simple matrix multiplication, so that the calculation speed is higher, and the parallelization processing is easier.
(3) The extracted session interest utilizes a Bi-GRU inter-session interest interaction layer to model a sequence conversion relation between sessions; and different weights are given to the extracted session features, so that the interest of the user in each session is considered, the session interest of the user is fused with the context information, the method has expressive force, and the recommendation performance is greatly improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the application and not to limit the invention.
FIG. 1 is a flowchart illustrating a method for interest recommendation based on user sequence click behavior according to an embodiment;
FIG. 2 is a block diagram of an interest recommendation model in accordance with one embodiment;
FIG. 3 is a diagram illustrating a structure of an interest extraction layer in a transform session according to an embodiment;
FIG. 4 is a block diagram of an interest recommendation system based on a user sequence click behavior according to an embodiment.
Detailed Description
The invention is further described with reference to the following figures and examples.
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Example one
Fig. 1 is a flowchart of an interest recommendation method based on a user sequence click behavior according to this embodiment. According to the interest recommendation method, firstly, the interaction sequences of the users are divided into different conversations according to the similarity of the preferences of the users in a certain time period, then the sequence recommendation is realized by combining the user interest in each conversation and the interest interaction among different conversations, and the problems that the existing sequence recommendation method ignores the internal structure of the user sequence behaviors and ignores the conversion relation among items are effectively solved.
Referring to fig. 1, the interest recommendation method based on user sequence click behavior includes the following steps:
s101, acquiring a historical interactive item sequence of the user.
Specifically, historical interaction item data of a user is obtained, and a historical interaction item sequence of the user is formed, wherein the historical interaction sequence of the user is represented by S ═ { h ═ h1,h2,h3...hn...hNAnd N is the total number of the user history interaction items.
In this embodiment, the interactive item data may be click item data of a user at a shopping website, song data listened to by the user at a music playing software, or movie data watched by the user at a video playing software.
S102, an interest recommendation model is constructed.
FIG. 2 is a schematic diagram of a structure of an interest recommendation model. As shown in FIG. 2, the interest recommendation model comprises a session division layer, an embedding layer, a transform intra-session interest extraction layer, a Bi-GRU inter-session interest interaction layer, an activation layer and a prediction layer.
And S103, session division.
Specifically, a historical interactive item sequence of a user is input into a session division layer of an interest recommendation model, and the session division layer of the interest recommendation model is used for dividing different sessions according to different items and different time spans, for example, the sessions are divided by a time interval of 30 minutes when the user clicks an item on a certain shopping website; or dividing the conversation by the time interval of 1 week when the user listens to the songs by certain music playing software; or divide the session with time intervals of one week when the user watches a movie on certain video playing software. The divided kth session may be denoted as Qk ═ h1,h2,h3…hiAnd f, wherein i is the total number of user interaction items in the kth session. After the session is divided, the mth item in the kth session can be represented as hkm。
In this embodiment, each item in each session has three identifiers, which are respectively the sequence feature identifier of the item, the position identifier of the item in the user click sequence, and the position identifier of the item in the session.
In this embodiment, after the historical interactive item sequence of the user is different sessions, the obtained sparse features of the user behavior and the target item in the session are input to the embedding layer of the interest recommendation model, because the embedding layer does not contain any convolution or cyclic structure, two learnable position embedding needs to be injected in the input process, the sparse features of the user behavior and the target item in the session are converted into low-dimensional dense features through the embedding layer, and the expression is as follows:
wherein, H is an item characteristic embedding matrix, P is an item position embedding matrix of the session where the item is located, and P is an item embedding matrix in the session. Through the embedding layer, a low-dimensional embedding vector Q can be obtained.
Past interest recommendation systems have generally used user-item interaction pairs as the basic unit of operation, in which case the user's intent at a certain point in time is easily overwhelmed by his historical behavior. In view of the above disadvantages, the present embodiment divides the click sequence behavior of the user into different sessions, and takes the session as the smallest operation unit, so as to reduce the information loss caused by ignoring or destroying the session structure to the maximum; and the conversation is used as a basic operation unit, which is more beneficial to parallel processing, and the relative method is very efficient in training efficiency.
And S104, extracting interest in the session.
Specifically, the interests of the users in a single session are similar, for example, if the user browses black hat and black shoes in one session, then it can be inferred that the interest preference of the user in the session is black clothes, so the embodiment inputs the low-dimensional dense feature vectors of the user behaviors and the target items from the embedded layer into the interest extraction layer in the transform session based on the multi-head self-attention mechanism, and extracts the interest of the user in each session.
FIG. 3 is a schematic diagram of the structure of an interest extraction layer in a transform session. As shown in FIG. 3, the low-dimensional dense feature vector Q obtained by the embedding layer first passes through a Multi-Head orientation layer in the interest extraction layer in the transform session, and the Multi-Head orientation can capture the relation of the items in different representation spaces. The Multi-Head Attenttion first projects vector Q linearly into h subspaces (h being the number of heads), then applies h Attention functions in parallel to produce an output list, the output of the ith Head in the kth session being:
headki=Attention(QkiWi Q,QkiWi K,QkiWi V)
wherein the content of the first and second substances,for a linear matrix, the Attention function in Multi-Head Attention is the scaled dot product Attention:
then, to impart a model non-linear relationship, the present embodiment connects the output structures of the different heads and inputs it into a feed-forward network:
FFN(x)=ReLu(xW(1)+b(1))+W(2)+b(2)
where S is the total number of items in session k and ReLU (. cndot.) is a non-linear activation function.
In summary, the session interest of the kth session can be expressed as:
wherein Avg (. cndot.) is the average pooling.
In addition, the present embodiment uses residual concatenation to reduce the complexity of the model and thus improve the performance of the attention mechanism.
After passing through the above-mentioned interest extraction layer in the transform session, the interaction sequence of the user can be further expressed as I ═ { I ═ I1,I2,……Ik}。
The transform intra-session interest extraction layer based on the multi-head self-attention mechanism can not only extract the interest of the user in each session, but also capture the intrinsic connection between user behaviors in the same session. In order to give the model non-linear characteristics, after connecting vectors of multiple heads from different heads of attention, the connecting result is fed into a feedforward network.
After the interaction sequence of the user is divided into different sessions, the interest of the user in each session is extracted by using an interest extraction layer in the multi-head self-attention-based transform session. Compared with the traditional modeling of interactive sequences through an RNN structure, the relevance of the interest extraction layer in the multi-head self-attention-based Transformer session does not depend on the distance and the distance of the current item and other items regardless of the spatial distance of the current item and other items, so that the sequence items can be better modeled. Meanwhile, the attention mechanism is only calculated by simple matrix multiplication, so that the calculation speed is higher, and the parallelization processing is easier.
And S105, interest interaction between the conversations.
Although the interests of the users are different in different sessions, the different interests hidden in different sessions are mutually influenced and correlated, for example, after the users buy the mobile phones in the previous session behaviors, the users are likely to browse the mobile phone shells next, so the embodiment extracts the interest in i sessions from step 103 and uses the Bi-GRU session interest interaction layer to model the dynamic interaction and conversion relationship among different sessions so as to capture the interaction and evolution of the interests of the users in different historical sessions.
GRUs, as variants of RNN and LSTM, have significant advantages in sequence modeling due to simpler structure. The formula for the GRU is as follows:
Here, in order to more accurately consider the influence of the context on the session, the present embodiment uses a Bi-GRU session interest interaction layer, whose expression is:
wherein the content of the first and second substances,is a hidden state of the forward GRU,is a hidden layer state of the backward GRU,are connection symbols.
And S106, activating.
Specifically, different session interests have different effects on whether the user clicks on the target item, so the user assigns different weights to the different session interests. The attention mechanism has been proven to be an effective weight allocation mechanism, so the embodiment utilizes the attention-based activation layer to assign different weights to different session interests so as to improve the accuracy of western region recommendation.
In this embodiment, the expression of the session interest weight is:
wherein α is a weighting factor that determines which session interest needs to be emphasized or ignored, UkAnd outputting the hidden layer vector after passing through the Bi-GRU session interest interaction layer.
After passing through the above interest interaction layer between Bi-GRU sessions, the interaction sequence of the user may be further expressed as U ═ U1,U2,……Uk}。
S107, predicting a target item sequence to be recommended.
Specifically, the interaction sequence I of the user obtained in step 104 and the interaction sequence U of the user obtained in step 106 are accessed to a prediction layer, and an n-dimensional representation of a next target item sequence to be recommended is obtained through prediction.
According to the method, different weights are given to the extracted session interests, the method not only considers the interests of the user in each session, but also fuses the session interests of the user with the context information, so that the method has expressive force and greatly improves the recommendation performance.
Example two
Fig. 4 is a block diagram of an interest recommendation system based on a user sequence click behavior according to the present embodiment. As shown in fig. 4, the system includes:
the data acquisition module is used for acquiring historical interactive project data of the user to form a historical interactive project sequence of the user;
the model construction module is used for constructing an interest recommendation model;
the session division module is used for carrying out session division on the historical interactive item sequence of the user by utilizing the interest recommendation model;
the intra-session interest extraction module is used for extracting each intra-session interest obtained after division;
the activation module is used for endowing different weights to the interests in each session to obtain a session interest sequence of the user;
the inter-session interest interaction module is used for interacting the interests among different sessions to obtain a dynamic interaction model among the different sessions;
and the prediction module is used for inputting the session interest sequence of the user into a dynamic interaction model among different sessions to predict and obtain a target item sequence to be recommended.
EXAMPLE III
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of:
acquiring historical interactive project data of a user to form a historical interactive project sequence of the user;
constructing an interest recommendation model;
carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model;
extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user;
interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations;
and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
Example four
The present embodiment provides a processing apparatus, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the processor implements the following steps:
acquiring historical interactive project data of a user to form a historical interactive project sequence of the user;
constructing an interest recommendation model;
carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model;
extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user;
interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations;
and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
As will be appreciated by one skilled in the art, embodiments of the present disclosure may be provided as a method, system, or computer program product. Accordingly, the present disclosure may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, and the like) having computer-usable program code embodied therein.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (10)
1. An interest recommendation method based on user sequence click behaviors is characterized by comprising the following steps:
acquiring historical interactive project data of a user to form a historical interactive project sequence of the user;
constructing an interest recommendation model;
carrying out session division on the historical interactive item sequence of the user by using an interest recommendation model;
extracting the interests in each session obtained after the division, and performing weighting processing on the interests in each session to obtain a session interest sequence of the user;
interacting the interests among different conversations to obtain a dynamic interaction model among the different conversations;
and inputting the session interest sequence of the user into a dynamic interaction model among different sessions, and predicting to obtain a target item sequence to be recommended.
2. The method of claim 1, wherein the interest recommendation model comprises a session division layer, an embedding layer, an intra-session interest extraction layer, an inter-session interest interaction layer, an activation layer, and a prediction layer.
3. The interest recommendation method based on user sequence click behavior as claimed in claim 1, wherein the step of session partitioning the historical interactive item sequence of the user comprises:
inputting the obtained historical interactive item sequence of the user into a session division layer of an interest recommendation model, dividing the items in the historical interactive item sequence of the user into a plurality of different sessions through the session division layer according to different time spans, wherein each item in the sessions has three identifiers which are respectively a sequence characteristic identifier of the item, a position identifier of the item in a user click sequence and a position identifier of the item in the session;
and inputting the obtained features of the items in the conversation into an embedding layer of the interest recommendation model, and obtaining the low-dimensional features of the items in the conversation through the embedding layer.
4. The interest recommendation method based on user sequence click behavior according to claim 1, wherein the step of extracting the interest in each session obtained after the division comprises:
inputting the low-dimensional feature vector of the item in the conversation output by the embedding layer of the interest recommendation model into the interest extraction layer in the conversation of the interest recommendation model;
and extracting the interest of the user in each session through an intra-session interest extraction layer.
5. The method of claim 4, wherein the step of weighting the interests in each session comprises:
and inputting the interest in each conversation into an activation layer of the interest recommendation model, and giving different weights to different conversation interests by using the activation layer to obtain a conversation interest sequence of the user.
7. The method of claim 1, wherein the step of interacting interests between different sessions comprises:
inputting the extracted interest of the user in each session into a session interest interaction layer of an interest recommendation model;
and modeling the dynamic interaction and conversion relation among different conversations by utilizing the conversation interest interaction layer.
8. An interest recommendation system based on user sequence click behaviors is characterized by comprising:
the data acquisition module is used for acquiring historical interactive project data of the user to form a historical interactive project sequence of the user;
the model construction module is used for constructing an interest recommendation model;
the session division module is used for carrying out session division on the historical interactive item sequence of the user by utilizing the interest recommendation model;
the intra-session interest extraction module is used for extracting intra-session interests obtained after division;
the activation module is used for endowing different weights to the interests in each session to obtain a session interest sequence of the user;
the inter-session interest interaction module is used for interacting the interests among different sessions to obtain a dynamic interaction model among the different sessions;
and the prediction module is used for inputting the session interest sequence of the user into a dynamic interaction model among different sessions to predict and obtain a target item sequence to be recommended.
9. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method for interest recommendation based on user sequence click behavior according to any one of claims 1 to 7.
10. A processing apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for interest recommendation based on user sequence click behavior according to any of claims 1-7 when executing the program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911012817.XA CN110807156A (en) | 2019-10-23 | 2019-10-23 | Interest recommendation method and system based on user sequence click behaviors |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911012817.XA CN110807156A (en) | 2019-10-23 | 2019-10-23 | Interest recommendation method and system based on user sequence click behaviors |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110807156A true CN110807156A (en) | 2020-02-18 |
Family
ID=69488987
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911012817.XA Pending CN110807156A (en) | 2019-10-23 | 2019-10-23 | Interest recommendation method and system based on user sequence click behaviors |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110807156A (en) |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111294619A (en) * | 2020-02-28 | 2020-06-16 | 电子科技大学 | Long-short term interest modeling method for IPTV field |
CN111382361A (en) * | 2020-03-12 | 2020-07-07 | 腾讯科技(深圳)有限公司 | Information pushing method and device, storage medium and computer equipment |
CN111723306A (en) * | 2020-05-29 | 2020-09-29 | 中国科学技术大学智慧城市研究院(芜湖) | Sequence position recommendation method based on geographical perception |
CN111737578A (en) * | 2020-06-22 | 2020-10-02 | 陕西师范大学 | Recommendation method and system |
CN111797321A (en) * | 2020-07-07 | 2020-10-20 | 山东大学 | Personalized knowledge recommendation method and system for different scenes |
CN111932336A (en) * | 2020-07-17 | 2020-11-13 | 重庆邮电大学 | Commodity list recommendation method based on long-term and short-term interest preference |
CN112115352A (en) * | 2020-08-28 | 2020-12-22 | 齐鲁工业大学 | Session recommendation method and system based on user interests |
CN112150206A (en) * | 2020-09-28 | 2020-12-29 | 京东数字科技控股股份有限公司 | Method and equipment for predicting user interested article |
CN112862007A (en) * | 2021-03-29 | 2021-05-28 | 山东大学 | Commodity sequence recommendation method and system based on user interest editing |
CN112905886A (en) * | 2021-02-22 | 2021-06-04 | 中国计量大学 | Session recommendation method based on multi-interest repeated network |
CN113360816A (en) * | 2020-03-05 | 2021-09-07 | 北京沃东天骏信息技术有限公司 | Click rate prediction method and device |
CN113495991A (en) * | 2020-04-03 | 2021-10-12 | 北京沃东天骏信息技术有限公司 | Recommendation method and device |
CN113609394A (en) * | 2021-08-09 | 2021-11-05 | 上海交通大学 | Information flow-oriented safety recommendation system |
CN113821724A (en) * | 2021-09-23 | 2021-12-21 | 湖南大学 | Graph neural network recommendation method based on time interval enhancement |
CN114154071A (en) * | 2021-12-09 | 2022-03-08 | 电子科技大学 | Emotion time sequence recommendation method based on attention mechanism |
CN114443956A (en) * | 2022-01-14 | 2022-05-06 | 腾讯科技(深圳)有限公司 | Content recommendation method and related equipment |
CN114791983A (en) * | 2022-04-13 | 2022-07-26 | 湖北工业大学 | Sequence recommendation method based on time sequence article similarity |
CN116521971A (en) * | 2022-01-19 | 2023-08-01 | 腾讯科技(深圳)有限公司 | Content recommendation method, apparatus, device, storage medium, and computer program product |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130091088A1 (en) * | 2011-10-07 | 2013-04-11 | George Forman | Making a recommendation to a user that is currently generating events based on a subset of historical event data |
CN110008409A (en) * | 2019-04-12 | 2019-07-12 | 苏州市职业大学 | Based on the sequence of recommendation method, device and equipment from attention mechanism |
CN110060097A (en) * | 2019-04-01 | 2019-07-26 | 苏州市职业大学 | User behavior sequence of recommendation method based on attention mechanism and convolutional neural networks |
CN110119467A (en) * | 2019-05-14 | 2019-08-13 | 苏州大学 | A kind of dialogue-based item recommendation method, device, equipment and storage medium |
-
2019
- 2019-10-23 CN CN201911012817.XA patent/CN110807156A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130091088A1 (en) * | 2011-10-07 | 2013-04-11 | George Forman | Making a recommendation to a user that is currently generating events based on a subset of historical event data |
CN110060097A (en) * | 2019-04-01 | 2019-07-26 | 苏州市职业大学 | User behavior sequence of recommendation method based on attention mechanism and convolutional neural networks |
CN110008409A (en) * | 2019-04-12 | 2019-07-12 | 苏州市职业大学 | Based on the sequence of recommendation method, device and equipment from attention mechanism |
CN110119467A (en) * | 2019-05-14 | 2019-08-13 | 苏州大学 | A kind of dialogue-based item recommendation method, device, equipment and storage medium |
Non-Patent Citations (1)
Title |
---|
高扬: "《人工智能与机器人先进技术丛书 智能摘要与深度学习》", 北京理工大学出版社 * |
Cited By (27)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111294619A (en) * | 2020-02-28 | 2020-06-16 | 电子科技大学 | Long-short term interest modeling method for IPTV field |
CN111294619B (en) * | 2020-02-28 | 2021-09-10 | 电子科技大学 | Long-short term interest modeling method for IPTV field |
CN113360816A (en) * | 2020-03-05 | 2021-09-07 | 北京沃东天骏信息技术有限公司 | Click rate prediction method and device |
CN111382361A (en) * | 2020-03-12 | 2020-07-07 | 腾讯科技(深圳)有限公司 | Information pushing method and device, storage medium and computer equipment |
CN111382361B (en) * | 2020-03-12 | 2023-05-02 | 腾讯科技(深圳)有限公司 | Information pushing method, device, storage medium and computer equipment |
CN113495991A (en) * | 2020-04-03 | 2021-10-12 | 北京沃东天骏信息技术有限公司 | Recommendation method and device |
CN111723306B (en) * | 2020-05-29 | 2023-07-25 | 长三角信息智能创新研究院 | Sequence position recommending method based on geographic perception |
CN111723306A (en) * | 2020-05-29 | 2020-09-29 | 中国科学技术大学智慧城市研究院(芜湖) | Sequence position recommendation method based on geographical perception |
CN111737578B (en) * | 2020-06-22 | 2024-04-02 | 陕西师范大学 | Recommendation method and system |
CN111737578A (en) * | 2020-06-22 | 2020-10-02 | 陕西师范大学 | Recommendation method and system |
CN111797321B (en) * | 2020-07-07 | 2021-04-27 | 山东大学 | Personalized knowledge recommendation method and system for different scenes |
CN111797321A (en) * | 2020-07-07 | 2020-10-20 | 山东大学 | Personalized knowledge recommendation method and system for different scenes |
CN111932336A (en) * | 2020-07-17 | 2020-11-13 | 重庆邮电大学 | Commodity list recommendation method based on long-term and short-term interest preference |
CN112115352A (en) * | 2020-08-28 | 2020-12-22 | 齐鲁工业大学 | Session recommendation method and system based on user interests |
CN112115352B (en) * | 2020-08-28 | 2024-07-05 | 齐鲁工业大学 | Session recommendation method and system based on user interests |
CN112150206A (en) * | 2020-09-28 | 2020-12-29 | 京东数字科技控股股份有限公司 | Method and equipment for predicting user interested article |
CN112905886A (en) * | 2021-02-22 | 2021-06-04 | 中国计量大学 | Session recommendation method based on multi-interest repeated network |
CN112862007B (en) * | 2021-03-29 | 2022-12-13 | 山东大学 | Commodity sequence recommendation method and system based on user interest editing |
CN112862007A (en) * | 2021-03-29 | 2021-05-28 | 山东大学 | Commodity sequence recommendation method and system based on user interest editing |
CN113609394B (en) * | 2021-08-09 | 2023-09-05 | 上海交通大学 | Information flow-oriented safety recommendation system |
CN113609394A (en) * | 2021-08-09 | 2021-11-05 | 上海交通大学 | Information flow-oriented safety recommendation system |
CN113821724B (en) * | 2021-09-23 | 2023-10-20 | 湖南大学 | Time interval enhancement-based graph neural network recommendation method |
CN113821724A (en) * | 2021-09-23 | 2021-12-21 | 湖南大学 | Graph neural network recommendation method based on time interval enhancement |
CN114154071A (en) * | 2021-12-09 | 2022-03-08 | 电子科技大学 | Emotion time sequence recommendation method based on attention mechanism |
CN114443956A (en) * | 2022-01-14 | 2022-05-06 | 腾讯科技(深圳)有限公司 | Content recommendation method and related equipment |
CN116521971A (en) * | 2022-01-19 | 2023-08-01 | 腾讯科技(深圳)有限公司 | Content recommendation method, apparatus, device, storage medium, and computer program product |
CN114791983A (en) * | 2022-04-13 | 2022-07-26 | 湖北工业大学 | Sequence recommendation method based on time sequence article similarity |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110807156A (en) | Interest recommendation method and system based on user sequence click behaviors | |
Li et al. | Spatio-temporal attention networks for action recognition and detection | |
Gabeur et al. | Multi-modal transformer for video retrieval | |
Ferreira et al. | Learning to dance: A graph convolutional adversarial network to generate realistic dance motions from audio | |
Wang et al. | Dualvgr: A dual-visual graph reasoning unit for video question answering | |
CN111339415B (en) | Click rate prediction method and device based on multi-interactive attention network | |
Tao et al. | Multi-modal knowledge-aware reinforcement learning network for explainable recommendation | |
CN110839173A (en) | Music matching method, device, terminal and storage medium | |
CN110364146A (en) | Audio recognition method, device, speech recognition apparatus and storage medium | |
CN112989212B (en) | Media content recommendation method, device and equipment and computer storage medium | |
CN117836765A (en) | Click prediction based on multimodal hypergraph | |
CN112650842A (en) | Human-computer interaction based customer service robot intention recognition method and related equipment | |
CN111954087B (en) | Method and device for intercepting images in video, storage medium and electronic equipment | |
CN115618101A (en) | Streaming media content recommendation method and device based on negative feedback and electronic equipment | |
CN112364184A (en) | Method, device, server and storage medium for ordering multimedia data | |
CN116051688A (en) | Transition animation generation method and device, computer readable storage medium and terminal | |
Zhu et al. | Describing unseen videos via multi-modal cooperative dialog agents | |
CN114896510A (en) | Intelligent news recommendation method and system based on user multi-interest characteristics | |
CN111310025A (en) | Model training method, data processing method, device and related equipment | |
CN113220929B (en) | Music recommendation method based on time residence and state residence mixed model | |
CN115482021A (en) | Multimedia information recommendation method and device, electronic equipment and storage medium | |
CN113542797A (en) | Interaction method and device in video playing and computer readable storage medium | |
Wang et al. | Efficient music analysis mechanism based on AI and IoT data mining | |
CN114647787A (en) | User personalized recommendation method based on multi-modal data | |
CN116263794A (en) | Double-flow model recommendation system and algorithm with contrast learning enhancement |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200218 |
|
RJ01 | Rejection of invention patent application after publication |