CN114637909A - Film recommendation system and method based on improved deep structured semantic model - Google Patents
Film recommendation system and method based on improved deep structured semantic model Download PDFInfo
- Publication number
- CN114637909A CN114637909A CN202210132815.XA CN202210132815A CN114637909A CN 114637909 A CN114637909 A CN 114637909A CN 202210132815 A CN202210132815 A CN 202210132815A CN 114637909 A CN114637909 A CN 114637909A
- Authority
- CN
- China
- Prior art keywords
- user
- movie
- vector
- module
- data
- 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 23
- 239000013598 vector Substances 0.000 claims abstract description 164
- 230000006399 behavior Effects 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000012545 processing Methods 0.000 claims abstract description 25
- 238000005516 engineering process Methods 0.000 claims abstract description 11
- 238000012163 sequencing technique Methods 0.000 claims abstract 2
- 230000009467 reduction Effects 0.000 claims description 27
- 230000005284 excitation Effects 0.000 claims description 22
- 238000007906 compression Methods 0.000 claims description 20
- 230000006835 compression Effects 0.000 claims description 20
- 238000000605 extraction Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 15
- 238000004140 cleaning Methods 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 230000002452 interceptive effect Effects 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 6
- 238000013144 data compression Methods 0.000 claims description 6
- 239000000284 extract Substances 0.000 claims description 5
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000007476 Maximum Likelihood Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000005065 mining Methods 0.000 claims description 3
- 230000003993 interaction Effects 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 235000021167 banquet Nutrition 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 239000003153 chemical reaction reagent Substances 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000012216 screening Methods 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
- G06F16/9535—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
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- 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
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/283—Multi-dimensional databases or data warehouses, e.g. MOLAP or ROLAP
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/783—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/70—Information retrieval; Database structures therefor; File system structures therefor of video data
- G06F16/78—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/7867—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, title and artist information, manually generated time, location and usage information, user ratings
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Fuzzy Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Quality & Reliability (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
A movie recommendation system based on an improved deep structured semantic model comprises a user behavior acquisition and processing module, an offline training module and an online recall and sorting module, wherein the user behavior acquisition and processing module is used for collecting interaction behaviors of users, the offline training module is used for receiving combined data output by the user behavior acquisition and processing module, the online recall and sorting module is used for taking out user feature vectors obtained by training of the users according to attribute features of the users, and recalling movie subsets recommended for the users in a movie vector library by adopting an approximate nearest neighbor search technology; a movie recommendation method based on an improved deep structured semantic model comprises the steps of user behavior acquisition and processing, offline training, online recall, sequencing and the like.
Description
Technical Field
The invention relates to the field of semantic models, in particular to a film recommendation system and method based on an improved deep structured semantic model.
Background
With the continuous development of the Internet, the network video market scale is increased year by year. Network users are constantly impacted by a large amount of redundant and invalid information while enjoying various video banquet with rich forms and contents. The huge data information far exceeds the bearable degree of the user, the correct selection of the user to the required information is seriously disturbed, the utilization rate of the information is very low, and even the user is troubled and disliked.
The recommendation system has been rapidly developed over the years as an effective means of solving the problem of "information overload" and its role in internet services is increasing. Movies, as a carrier of rich information, naturally become an important research object in personalized recommendations. With the increasing number of users and movies, how to deeply mine movie information and accurately match interests of the users, a suitable movie is selected for the users from the vast movie library, and accurate personalized services are provided, which has become a hotspot of research in the industry.
The movie recommendation algorithm is the core of a movie recommendation system, and currently, a relatively large number of applications are collaborative filtering, which is to find similar users who have watched the same video as a target user based on the watching history of the user, and then find other videos that the similar users like to watch, and recommend the videos to the target user. The problems with this solution are: when the scoring records of the movie by the user are very few, the collaborative filtering is used for recommending the poor effect, namely the cold start problem; with the increasing of users and movies, the collaborative filtering matrix is very large, and the cost of obtaining the feature vectors of the users and the movies by matrix decomposition is higher and higher. Other solutions are to learn the historical preferences of the user, such as using a deep neural network or a recurrent neural network. At present, the research of the movie recommendation algorithm has a large development space.
Disclosure of Invention
The invention aims to provide a movie recommendation system and method based on an improved deep structured semantic model, so as to solve the problems in the prior art.
In order to achieve the above object, the present invention provides a movie recommendation system based on an improved deep-structured semantic model, which comprises a user behavior acquisition and processing module, an offline training module and an online recall and ranking module,
the user behavior acquisition and processing module collects an interactive behavior log, a search behavior log and a play record list of a user by embedding points at the front end, stores the interactive behavior log, the search behavior log and the play record list as user characteristic data into a file system characteristic library, and performs data cleaning on the collected log by means of a data warehouse tool to obtain a basic sample original data set;
according to the original data obtained by cleaning, the system preprocesses the behavior logs of the users and performs data combination on the behavior logs of the users;
the offline training module receives the combined data sample output by the user behavior acquisition and processing module, reweighs the data after coding and dimensionality reduction of the data, extracts and mines deep latent semantic features of the data, and matches the user with the movie according to the data features;
and the online recall and sorting module is used for taking out the user characteristic vector obtained by the training of the user according to the attribute characteristics of the user, performing vector retrieval in a movie vector library by adopting an approximate nearest neighbor search technology, and recalling the movie subset recommended for the user.
The off-line training module comprises an input layer, a self-attention layer, a feature extraction layer and a matching layer,
the user characteristic data output by the user behavior acquisition and processing module is merged and then sent to an input layer, the input layer comprises a coding module and a dimension reduction module, and the input layer inputs the user characteristic data to the coding module and the dimension reduction module;
the self-attention layer re-weights the data of the input layer by adopting a compressed excitation network SENET;
the feature extraction layer uses three fully-connected networks to form a deep neural network for extracting and mining deep latent semantic features of input user and movie feature vectors;
and the matching layer obtains the matching score between the user and the movie by calculating the cosine similarity between the extracted implicit feature vectors according to the extracted implicit feature vectors.
The user feature data comprise user dense features and user sparse features, the user dense features are input to the coding module, and the user sparse features are input to the dimension reduction module;
the user sparse features comprise sparse features with determined values and variable-length sparse features, and the sparse features with the determined values are input into the coding module and then output as low-dimensional vectors;
the variable-length sparse features comprise viewing history and search history, and the movie embedding sequences corresponding to the viewing history are subjected to vector weighted average by a dimensionality reduction module to obtain viewing vectors;
the search history is trained by a dimensionality reduction module to obtain an embedded vector, and the corresponding film embedded sequence is weighted and averaged to obtain a search vector;
and the search history and the viewing history are trained in a staggered and corresponding mode, the input layer splices the processed sparse features and the processed dense features, and the spliced user and movie vectors are used as initial embedded vectors.
The self-attention layer comprises a compression module and an excitation module, and the compression module performs data compression and information summarization on the embedded vector of each feature received from the input layer to form an initial weight vector;
the excitation module is used for performing feature crossing on the initial weight vector output by the compression module and keeping the dimension of output size;
and the offline recommendation module matching layer calculates cosine similarity between the implicit semantic feature vectors extracted by the feature extraction layer according to the implicit semantic feature vectors extracted by the feature extraction layer to obtain a matching score between the user and the film.
The online recall and sorting module takes out the user characteristic vector trained by the user according to the attribute characteristics of the user, recalls the film subset recommended for the user in the film vector library by adopting an approximate nearest neighbor search technology, removes the films watched by the user in a sorting stage, calculates the similarity between the residual films and the user characteristic vector, takes the similarity as a sorting basis, and returns a recommendation result list.
A movie recommendation method based on an improved deep structured semantic model comprises the following steps:
s1: user behavior acquisition and processing: the method comprises the steps that interactive behavior logs, search behavior logs and play record lists of users are collected through point burying at the front end and stored in a file system, and data cleaning is conducted on the collected logs by means of a data warehouse tool to obtain an original data set;
according to the original data obtained by cleaning, the system preprocesses the behavior logs of the users and performs data combination on the behavior logs of the users;
s2: off-line training: the combined data output by the user behavior acquisition and processing module is received by S1, the data is re-weighted after being encoded and dimensionality reduced, deep latent semantic features of the data are extracted and mined, and the user and the movie are matched according to the data features;
s3: online recall and ranking: and extracting the user feature vector trained by the user according to the attribute features of the user, and recalling the film subset recommended for the user in a film vector library by adopting an approximate nearest neighbor search technology.
In step S2, the method further includes:
a1: characteristic input: after the combined data output by the user behavior acquisition and processing is received, encoding and dimension reduction are carried out on the user characteristic data;
a2: and (3) feature learning: re-weighting the data of the input layer by adopting a compressed excitation network SENET;
a3: characteristic extraction: three full-connection networks are used for forming a deep neural network, and deep latent semantic features of input users and movie feature vectors are extracted and mined;
a4: and (3) feature matching: and according to the extracted latent semantic feature vectors, calculating the cosine similarity between the latent semantic feature vectors to obtain the matching score between the user and the movie.
In step a1, the user sparse features include a determined value sparse feature and a variable length sparse feature, and the user sparse features are encoded and then output as low-dimensional vectors;
the variable-length sparse features comprise viewing history and search history, the movie embedding sequence corresponding to the viewing history is subjected to dimension reduction vector weighted average to obtain a viewing vector, and the formula is as follows:
where t denotes the current time, t0Represents a viewing time, miIs the embedded vector for the ith movie;
the search history is subjected to dimensionality reduction training to obtain an embedded vector, the corresponding film embedded sequence is subjected to weighted average to obtain a search vector, and the formula is as follows:
and the search history and the viewing history are trained in a staggered and corresponding mode, the input layer splices the processed sparse features and the processed dense features, and the spliced user and movie vectors are used as initial embedded vectors.
Step a2, further including a compression stage and an excitation stage, where the compression stage performs data compression and information summarization on the embedded vector of each feature received from step a1 to form an initial weight vector, and the formula is as follows:
the excitation stage is used for performing characteristic crossing on the initial weight vector output in the compression stage and keeping the output size dimension, two layers of MLP networks with narrower middle layers are introduced in the compression stage and act on the output vector Z in the excitation stage, and the formula is as follows:
S＝Fex(Z,W)＝δ(W2δ(W1Z))；
where δ is the activation function, the first MLP functions to do the feature crossing, and the second MLP functions to maintain the size dimension of the output.
In step a3, a deep neural network is formed by using three fully-connected networks, deep latent semantic features are extracted and mined from input user and movie feature vectors, and a latent semantic feature vector y is specifically represented as:
li＝f(Wili-1+bi),i＝2,…,N-1；
y＝f(WNlN-1+bN)；
wherein, { liWhere i is 1,2, …, N-1, and W represents the output of each fully-connected layeri，biRespectively representing the weight matrix and the bias term of the ith layer,
f represents the activation function tanh:
step A4, extracting the extracted implicit characteristic vectors according to the characteristics of the step A3, and calculating the cosine similarity between the implicit characteristic vectors and the implicit characteristic vectors to obtain the matching score between the user and the movie;
wherein, yU、yMLatent semantic feature vectors representing the resulting user and movie, respectively, | | representing a modulo operation;
during model training, the cosine similarity of the final feature vectors of the user and the film is converted into posterior probability through a softmax function, and the formula is as follows:
wherein γ represents a smoothing factor of the softmax function, and minimizes a loss function through maximum likelihood estimation, and aims to maximize the user viewing duration by adding time weight, and the formula is as follows:
L(Λ)＝-log∏(U,M+)Tj·P(M+|U)；
wherein, TjRepresenting the duration of the j-th movie, M representing the set of candidate movies, Lambda representing the model parameters, M+Representing positive samples in the candidate movie.
Step S3, according to the attribute characteristics of the user, the user characteristic vector trained by the user is taken out, the movie subset recommended for the user is recalled in the movie vector library by adopting the approximate nearest neighbor search technology, the movies watched by the user are removed in the sorting stage, the similarity between the remaining movies and the user characteristic vector is calculated, and the similarity is used as the sorting basis, and the recommendation result list is returned.
Compared with the prior art, the invention has the beneficial effects that:
1. the improved deep structured semantic recommendation model can effectively dig out the movies according with the user interests according to the dominant characteristics and the hidden interaction information of the users and the movies, and provides personalized recommendation service for the users;
2. the method uses a Faiss frame to store the feature vectors of the user and the movie trained by the model according to the attribute features of the user and the movie, when the user accesses the system again, the vectors can be quickly taken out and ANN search can be carried out according to the feature vectors, so that the recommendation result of the user is obtained;
3. the movie recommendation system provided by the patent can continuously collect user behavior logs and regularly update the historical records into the portrait of the user, so that the system can learn the recent interest of the user in time.
Drawings
FIG. 1 is a schematic diagram of a movie recommendation system based on an improved deep-structured semantic model according to the present invention;
FIG. 2 is a diagram of an off-line training model structure of a movie recommendation system based on an improved deep-structured semantic model according to the present invention;
FIG. 3 is a schematic diagram of a movie recommendation system based on an improved deep-structured semantic model for solving a time crossing problem according to the present invention;
FIG. 4 is a flowchart of an embodiment of a movie recommendation method based on an improved deep-structured semantic model according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. The examples, in which specific conditions are not specified, were carried out according to conventional conditions or conditions recommended by the manufacturer. The reagents or instruments used are not indicated by the manufacturer, and are all conventional products available commercially.
Example (b): as shown in FIG. 1, a movie recommendation system based on an improved deep structured semantic model includes a user behavior acquisition and processing module, an offline training module and an online recall and ranking module,
the user behavior acquisition and processing module collects interactive behavior logs, search behavior logs and play record lists of users through embedding points at the front end, and stores the interactive behavior logs, the search behavior logs and the play record lists as user characteristic data into a file system characteristic library, the characteristic library of the embodiment adopts a Faiss frame, and data cleaning is carried out on the collected logs by means of a data warehouse tool to obtain a basic sample original data set;
according to the original data obtained by cleaning, the system preprocesses the behavior logs of the users and performs data combination on the behavior logs of the users;
the offline training module receives the combined data sample output by the user behavior acquisition and processing module, rewrites the data after coding and dimensionality reduction of the data, extracts and mines deep latent semantic features of the data to complete model training, matches the user with a movie according to the data features, and completes model prediction;
the online recall and sorting module extracts user feature vector model prediction data obtained by training of a user according to the attribute features of the user, and performs vector retrieval in a movie vector library by adopting an approximate nearest neighbor search technology, wherein in the embodiment, the vector retrieval adopts nearest neighbor search ANN to recall the movie subset recommended for the user.
As shown in fig. 2, the offline training module includes an input layer, a self-attention layer, a feature extraction layer and a matching layer,
the user characteristic data output by the user behavior acquisition and processing module are merged and then sent to an input layer, the input layer comprises a coding module and a dimension reduction module, and the input layer inputs the user characteristic data to the coding module and the dimension reduction module;
the self-attention layer adopts a compressed excitation network SENET to re-weight the data of the input layer;
the feature extraction layer uses three fully-connected networks to form a deep neural network for extracting and mining deep latent semantic features of input user and movie feature vectors;
and the matching layer calculates the cosine similarity between the extracted latent semantic feature vectors according to the extracted latent semantic feature vectors to obtain the matching score between the user and the film.
The user feature data comprise user dense features and user sparse features, the user dense features are input to the coding module, and the user sparse features are input to the dimension reduction module;
the user sparse features comprise sparse features with determined values and variable-length sparse features, and the sparse features with the determined values are input into the coding module and then output as low-dimensional vectors;
the variable-length sparse features comprise viewing history and search history, and a viewing vector is obtained after a movie Embedding sequence corresponding to the viewing history is subjected to vector weighted average by a dimensionality reduction module;
training the search history through a dimensionality reduction module to obtain an Embedding vector, and carrying out weighted average on the corresponding Embedding sequence of the movie to obtain a search vector;
and training the search history and the viewing history in a staggered and corresponding manner, splicing the processed sparse features and the processed dense features by the input layer, and using the spliced user and movie vectors as initial Embedding vectors.
The self-attention layer comprises a compression module and an excitation module, wherein the compression module is used for carrying out data compression and information summarization on the Embedding vector of each feature received from the input layer to form an initial weight vector;
the excitation module is used for performing feature crossing on the initial weight vector output by the compression module and keeping the dimension of output size;
and the offline recommendation module matching layer calculates the cosine similarity between the implicit semantic feature vectors extracted by the feature extraction layer according to the implicit semantic feature vectors extracted by the feature extraction layer to obtain the matching score between the user and the movie.
And the online recall and sorting module is used for taking out a user characteristic vector obtained by training of the user according to the attribute characteristics of the user, recalling the movie subset recommended for the user in the movie vector library by adopting an approximate nearest neighbor search technology ANN, removing movies already watched by the user in a sorting stage, calculating the similarity between the rest movies and the user characteristic vector, taking the similarity as a sorting basis, and returning a recommendation result list.
A movie recommendation method based on an improved deep structured semantic model comprises the following steps:
s1: user behavior collection and processing: the method comprises the steps that interactive behavior logs, search behavior logs and play record lists of users are collected through point burying at the front end and stored in a file system, and data cleaning is conducted on the collected logs through a data warehouse tool to obtain an original data set;
according to the original data obtained by cleaning, the system preprocesses the behavior logs of the users and performs data combination on the behavior logs of the users;
s2: off-line training: the combined data output by the user behavior acquisition and processing module is received by S1, the data is re-weighted after being encoded and dimensionality reduced, deep latent semantic features of the data are extracted and mined, and the user and the movie are matched according to the data features;
s3: online recall and ordering: and extracting the user feature vector obtained by the training of the user according to the attribute features of the user, and recalling the movie subset recommended for the user in the movie vector library by adopting an approximate nearest neighbor search technology.
In step S2, the method further includes:
a1: characteristic input: after receiving the combined data acquired by user behavior and processed and output, coding and reducing the dimension of the user characteristic data;
a2: and (3) feature learning: re-weighting the data of the input layer by adopting a compressed excitation network SENET;
a3: characteristic extraction: three full-connection networks are used for forming a deep neural network, and deep latent semantic features of input users and movie feature vectors are extracted and mined;
a4: and (3) feature matching: and according to the extracted latent semantic feature vectors, calculating the cosine similarity between the latent semantic feature vectors to obtain the matching score between the user and the movie.
In the step A1, the user sparse features comprise sparse features with determined values and variable-length sparse features, and the sparse features with the determined values are output as low-dimensional vectors after being coded;
performing one-hot coding operation on the dense features of the user; and carrying out embedding dimension reduction to a low-dimensional space operation on the sparse features of the user.
The processing of the user sparse features can be divided into two types:
one is to process sparse features with certain values: the characteristics mainly include coding characteristics such as user numbers, user ages, user occupation and the like, and each user of the characteristics only has a unique definite value. A word embedding model can be created by using a machine learning library such as a pyrrch and a torch.
The variable-length sparse features comprise viewing history and search history, the movie Embedding sequence corresponding to the viewing history is subjected to dimension reduction vector weighted average to obtain a viewing vector, and the formula is as follows:
where t denotes the current time, t0Represents a viewing time, miIs the Embedding vector of the ith movie;
the processing method of the user search history is similar to the processing method of the viewing history, firstly, the keywords of the history search are segmented to obtain an entry token, an Embedding vector of the token is obtained through training, then the Embedding vector corresponding to the token of the user history search is weighted and averaged to obtain a search vector, so that the overall state of the user search history can be learned, and the shorter the time interval with the current time is, the higher the weight of the Embedding vector is, and the formula is as follows:
the search history is subjected to dimensionality reduction training to obtain an embedded vector, the corresponding film embedded sequence is subjected to weighted average to obtain a search vector, and the formula is as follows:
for the training of the two variable-length sequence features, the system has a punishment mechanism for over-active users, and can set a sequence feature upper limit for each user, so that the model is prevented from being represented by a small number of over-active users, and is equally recommended for each user.
As shown in FIG. 3, considering that users watch videos and search frequently in a sequential manner, some front and back watching behaviors even have some causal associations, a traditional recommendation model is trained by taking front and back context as input information, namely disclosing some future information, in order to solve the time crossing problem, a training sample completely separates the future information, and the time crossing problem is solved by training by using a search record of T-2 and a viewing history record of T-1.
And training the search history and the viewing history in a staggered corresponding manner, splicing the processed sparse features and the processed dense features by the input layer, and taking the spliced user and movie vectors as initial embedded vectors.
In step a2, a compressed Excitation network SENET is used to dynamically learn the importance of these features, reduce the weight to suppress invalid low-frequency features, and increase the weight to amplify important features, so as to re-weight the initial embedded vector of users and movies, specifically including a Squeeze compression phase and an Excitation phase, where the Squeeze compression phase performs data compression and information summarization on the embedded vector of each feature received from step a1 to form an initial weight vector, and the following formula:
suppose a certain feature uiThe Embedding vector dimension of (a) is k, then we carry out the operation of averaging k numbers contained in the Embedding vector of (a) Embedding, and obtain the value z capable of representing the summary information of the featurei；
By the Squeeze compression stage, for each feature uiAll compressed to a single value ziAssuming that f features exist in the spliced Embedding vector, an initial weight vector Z can be formed as { Z ═ Z1,z2,…,zf}；
In the Excitation stage, two layers of MLP networks with narrower middle layers are introduced to act on an output vector Z in the Excitation stage, and the formula is as follows:
S＝Fex(Z,W)＝δ(W2δ(W1Z))；
where δ is the activation function, the first MLP acts to do the feature crossing, and the second MLP acts to maintain the size dimension of the output.
In step a3, a deep neural network is formed by using three full-connection networks DNN, deep latent semantic features are extracted and mined from input user and movie feature vectors, and a latent semantic feature vector y is specifically represented as:
li＝f(Wili-1+bi),i＝2,…,N-1；
y＝f(WNlN-1+bN)；
wherein, { liWhere i is 1,2, …, N-1, and W represents the output of each fully-connected layeri，biRespectively representing the weight matrix and the bias term of the ith layer,
f represents the activation function tanh:
step A4, extracting the extracted implicit characteristic vectors according to the characteristics of the step A3, and calculating the cosine similarity between the implicit characteristic vectors and the implicit characteristic vectors to obtain the matching score between the user and the movie;
wherein, yU、yMLatent semantic feature vectors representing the resulting user and movie, respectively, | | representing a modulo operation;
the input of the model is composed of a user and a set consisting of a plurality of candidate movies, the movie set comprises positive and negative samples with a certain proportion, wherein the division standard of the positive and negative proportion is according to the preference degree of the user to the movies, the preference degree is more than 0.5 and is positive sample, and the preference degree is less than 0.3 and is negative sample. The calculation formula of the preference degree is as follows:
wherein,
representing user uiFor movie mjIf the viewing duration is less than thirty percent of the movie duration, a score of 1 is obtained, if it is more than seventy percent, a score of 2 is obtained,
and
respectively represent users uiFor movie mjThe scores of the praise and the forward are set to be 3 points and 4 points.
During model training, the cosine similarity of the final feature vectors of the user and the film is converted into posterior probability through a softmax function, and the formula is as follows:
wherein γ represents a smoothing factor of the softmax function, and minimizes a loss function through maximum likelihood estimation, and by adding a time weight, the objective is to maximize the user viewing duration, and the formula is as follows:
wherein, TjRepresenting the duration of the j-th movie, M representing the set of candidate movies, Λ representing the model parameter, M+Representing positive samples in the candidate movie.
Step S3, according to the attribute characteristics of the user, the user characteristic vector which is obtained by the training of the user is taken out, the movie subset recommended for the user is recalled in the movie vector library by adopting the approximate nearest neighbor search technology, the movies which are already watched by the user are removed in the sorting stage, the similarity between the remaining movies and the user characteristic vector is calculated, and the rest movies are used as the sorting basis, and the recommendation result list is returned.
The specific implementation flow chart is shown in fig. 4. Firstly, a user enters the system to log in. The system judges whether the user is a new user, if not, the system indicates that the system has historical data, the system already obtains the feature vector of the user in the latent semantic space, and establishes an index according to the ID of the user and stores the index into a Faiss frame. The system takes the feature vector according to the user ID, and recalls the movie candidate set which is possibly interested by the user by performing approximate nearest neighbor search (ANN) in the movie vector library by using the feature vector. Because the movie candidate set may have movies already watched by the user, the system will perform screening on the movie candidate set again in the sorting stage to remove the movies already watched, calculate similarity between the remaining movies and the feature vectors of the user, and use the similarity as a sorting basis, and return a recommendation result list of TOP-N.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein, and any reference signs in the claims are not intended to be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.
Claims (10)
1. A movie recommendation system based on an improved deep structured semantic model is characterized by comprising a user behavior acquisition and processing module, an offline training module and an online recall and sequencing module,
the user behavior acquisition and processing module collects an interactive behavior log, a search behavior log and a play record list of a user by embedding points at the front end, stores the interactive behavior log, the search behavior log and the play record list as user characteristic data into a file system characteristic library, and performs data cleaning on the collected log by means of a data warehouse tool to obtain a basic sample original data set;
according to the original data obtained by cleaning, the system preprocesses the behavior logs of the users and performs data combination on the behavior logs of the users;
the offline training module receives the combined data sample output by the user behavior acquisition and processing module, reweighs the data after coding and dimensionality reduction of the data, extracts and mines deep latent semantic features of the data, and matches the user with the movie according to the data features;
and the online recall and sorting module is used for taking out the user characteristic vector obtained by the training of the user according to the attribute characteristics of the user, performing vector retrieval in a movie vector library by adopting an approximate nearest neighbor search technology, and recalling the movie subset recommended for the user.
2. The movie recommendation system based on the improved deep-structured semantic model according to claim 1,
the off-line training module comprises an input layer, a self-attention layer, a feature extraction layer and a matching layer,
the user characteristic data output by the user behavior acquisition and processing module is merged and then sent to an input layer, the input layer comprises a coding module and a dimension reduction module, and the input layer inputs the user characteristic data to the coding module and the dimension reduction module;
the self-attention layer re-weights the data of the input layer by adopting a compressed excitation network SENET;
the feature extraction layer uses three fully-connected networks to form a deep neural network for extracting and mining deep latent semantic features of input user and movie feature vectors;
and the matching layer calculates the cosine similarity between the extracted latent semantic feature vectors according to the extracted latent semantic feature vectors to obtain the matching score between the user and the film.
3. The movie recommendation system based on the improved deep-structured semantic model according to claim 2,
the user feature data comprise user dense features and user sparse features, the user dense features are input to the coding module, and the user sparse features are input to the dimension reduction module;
the user sparse features comprise sparse features with determined values and variable-length sparse features, and the sparse features with the determined values are input into the coding module and then output as low-dimensional vectors;
the variable-length sparse features comprise viewing history and search history, and the movie embedding sequences corresponding to the viewing history are subjected to vector weighted average by a dimensionality reduction module to obtain viewing vectors;
the search history is trained by a dimensionality reduction module to obtain an embedded vector, and the corresponding film embedded sequence is weighted and averaged to obtain a search vector;
and the search history and the viewing history are trained in a staggered and corresponding mode, the input layer splices the processed sparse features and the processed dense features, and the spliced user and movie vectors are used as initial embedded vectors.
4. The movie recommendation system based on the improved deep-structured semantic model according to claim 2,
the self-attention layer comprises a compression module and an excitation module, and the compression module performs data compression and information summarization on the embedded vector of each feature received from the input layer to form an initial weight vector;
the excitation module is used for performing feature crossing on the initial weight vector output by the compression module and keeping the dimension of output size;
and the offline recommendation module matching layer calculates cosine similarity between the implicit semantic feature vectors extracted by the feature extraction layer according to the implicit semantic feature vectors extracted by the feature extraction layer to obtain a matching score between the user and the film.
5. The movie recommendation system based on the improved deep-structured semantic model according to claim 1, wherein the online recall and sorting module extracts a user feature vector obtained by training a user according to attribute features of the user, recalls a subset of movies recommended for the user in a movie vector library by using an approximate nearest neighbor search technique, removes movies already watched by the user in a sorting stage, calculates similarity between the remaining movies and the user feature vector, and returns a recommendation result list by using the similarity as a sorting basis.
6. A movie recommendation method based on an improved deep structured semantic model is characterized by comprising the following steps:
s1: user behavior acquisition and processing: the method comprises the steps that interactive behavior logs, search behavior logs and play record lists of users are collected through point burying at the front end and stored in a file system, and data cleaning is conducted on the collected logs by means of a data warehouse tool to obtain an original data set;
according to the original data obtained by cleaning, the system preprocesses the behavior logs of the users and performs data combination on the behavior logs of the users;
s2: off-line training: the combined data output by the user behavior acquisition and processing module is received by S1, the data is re-weighted after being encoded and dimensionality reduced, deep latent semantic features of the data are extracted and mined, and the user and the movie are matched according to the data features;
s3: online recall and ranking: and extracting the user feature vector obtained by the training of the user according to the attribute features of the user, and recalling the movie subset recommended for the user in the movie vector library by adopting an approximate nearest neighbor search technology.
7. The method for recommending movies based on the improved deep-structured semantic model as claimed in claim 6, wherein in step S2, the method further comprises the following steps:
a1: characteristic input: after the combined data output by the user behavior acquisition and processing is received, encoding and dimension reduction are carried out on the user characteristic data;
a2: and (3) feature learning: re-weighting the data of the input layer by adopting a compressed excitation network SENET;
a3: characteristic extraction: three full-connection networks are used for forming a deep neural network, and deep latent semantic features of input users and movie feature vectors are extracted and mined;
a4: characteristic matching: and according to the extracted latent semantic feature vectors, calculating the cosine similarity between the latent semantic feature vectors to obtain the matching score between the user and the movie.
8. The movie recommendation method based on the improved deep-structured semantic model according to claim 7, wherein in step a1, the sparse features of the user include sparse features with definite values and variable-length sparse features, and the sparse features with definite values are encoded and then output as low-dimensional vectors;
the variable-length sparse features comprise viewing history and search history, the movie embedding sequence corresponding to the viewing history is subjected to dimension reduction vector weighted average to obtain a viewing vector, and the formula is as follows:
where t denotes the current time, t0Represents a viewing time, miIs the embedded vector for the ith movie;
the search history is subjected to dimensionality reduction training to obtain an embedded vector, the corresponding film embedded sequence is subjected to weighted average to obtain a search vector, and the formula is as follows:
and the search history and the viewing history are trained in a staggered and corresponding mode, the input layer splices the processed sparse features and the processed dense features, and the spliced user and movie vectors are used as initial embedded vectors.
9. The movie recommendation method based on the improved deep-structured semantic model according to claim 7, further comprising a compression stage and an excitation stage in step a2, wherein the compression stage performs data compression and information summarization on the embedded vector of each feature received in step a1 to form an initial weight vector, and the formula is as follows:
the excitation stage is used for performing feature crossing on the initial weight vector output in the compression stage and keeping the dimension of the output size, two layers of MLP networks with narrower middle layers are introduced in the compression stage and act on the output vector Z in the excitation stage, and the formula is as follows:
S＝Fex(Z,W)＝δ(W2δ(W1Z))；
where δ is the activation function, the first MLP acts to do the feature crossing, and the second MLP acts to maintain the size dimension of the output.
In step a3, a deep neural network is formed by using three fully-connected networks, deep latent semantic features are extracted and mined from input user and movie feature vectors, and a latent semantic feature vector y is specifically represented as:
li＝f(Wili-1+bi),i＝2,…,N-1；
y＝f(WNlN-1+bN)；
wherein, { liWhere i is 1,2, …, N-1, and W represents the output of each fully-connected layeri，biRespectively representing the weight matrix and the bias term of the ith layer,
f represents the activation function tanh:
step A4, extracting the extracted implicit characteristic vectors according to the characteristics of the step A3, and calculating the cosine similarity between the implicit characteristic vectors and the implicit characteristic vectors to obtain the matching score between the user and the movie;
wherein, yU、yMLatent semantic feature vectors representing the resulting user and movie, respectively, | | represents a modulo operation;
during model training, the cosine similarity of the final feature vectors of the user and the film is converted into posterior probability through a softmax function, and the formula is as follows:
wherein γ represents a smoothing factor of the softmax function, and minimizes a loss function through maximum likelihood estimation, and by adding a time weight, the objective is to maximize the user viewing duration, and the formula is as follows:
wherein, TjRepresenting the duration of the j-th movie, M representing the set of candidate movies, Lambda representing the model parameters, M+Representing positive samples in the candidate movie.
10. The method according to claim 6, wherein in step S3, the user feature vector obtained by training the user is extracted according to the attribute features of the user, the subset of movies recommended for the user is recalled in the movie vector library by using the approximate nearest neighbor search technique, the movies already watched by the user are removed in the sorting stage, the similarity between the remaining movies and the user feature vector is calculated, and the similarity is used as a sorting criterion, and the recommendation result list is returned.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210132815.XA CN114637909A (en) | 2022-02-14 | 2022-02-14 | Film recommendation system and method based on improved deep structured semantic model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210132815.XA CN114637909A (en) | 2022-02-14 | 2022-02-14 | Film recommendation system and method based on improved deep structured semantic model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114637909A true CN114637909A (en) | 2022-06-17 |
Family
ID=81946036
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210132815.XA Pending CN114637909A (en) | 2022-02-14 | 2022-02-14 | Film recommendation system and method based on improved deep structured semantic model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114637909A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE202023104110U1 (en) | 2023-07-23 | 2023-07-28 | Upasana Adhikari | Intelligent encryption-based system for movie recommendations |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180067935A1 (en) * | 2017-08-24 | 2018-03-08 | Prakash Kumar | Systems and methods for digital media content search and recommendation |
| CN110162706A (en) * | 2019-05-22 | 2019-08-23 | 南京邮电大学 | A kind of personalized recommendation method and system based on interaction data cluster |
| CN113051468A (en) * | 2021-02-22 | 2021-06-29 | 山东师范大学 | Movie recommendation method and system based on knowledge graph and reinforcement learning |
-
2022
- 2022-02-14 CN CN202210132815.XA patent/CN114637909A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180067935A1 (en) * | 2017-08-24 | 2018-03-08 | Prakash Kumar | Systems and methods for digital media content search and recommendation |
| CN110162706A (en) * | 2019-05-22 | 2019-08-23 | 南京邮电大学 | A kind of personalized recommendation method and system based on interaction data cluster |
| CN113051468A (en) * | 2021-02-22 | 2021-06-29 | 山东师范大学 | Movie recommendation method and system based on knowledge graph and reinforcement learning |
Non-Patent Citations (2)
| Title |
|---|
| XUGANG YE 等: "Enhancing Retrieval and Ranking Performance for Media Search Engine by Deep Learning", 2016 49TH HAWAII INTERNATIONAL CONFERENCE ON SYSTEM SCIENCES, 31 December 2016 (2016-12-31), pages 1 - 7 * |
| 常志 等: "基于深度学习的视频描述方法研究综述", 天津理工大学学报, vol. 36, no. 6, 31 December 2020 (2020-12-31), pages 1 - 7 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE202023104110U1 (en) | 2023-07-23 | 2023-07-28 | Upasana Adhikari | Intelligent encryption-based system for movie recommendations |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102334118B (en) | Promoting method and system for personalized advertisement based on interested learning of user | |
| CN112163122B (en) | Method, device, computing equipment and storage medium for determining label of target video | |
| CN105138653B (en) | It is a kind of that method and its recommendation apparatus are recommended based on typical degree and the topic of difficulty | |
| EP3367676A1 (en) | Video content analysis for automatic demographics recognition of users and videos | |
| CN112100504B (en) | Content recommendation method and device, electronic equipment and storage medium | |
| CN116157791A (en) | Text recommendation method, model training method, device and readable storage medium | |
| CN109034953B (en) | Movie recommendation method | |
| CN112464100B (en) | Information recommendation model training method, information recommendation method, device and equipment | |
| CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
| CN112364245B (en) | Top-K movie recommendation method based on heterogeneous information network embedding | |
| Deldjoo et al. | Using Visual Features and Latent Factors for Movie Recommendation. | |
| CN112364184A (en) | Method, device, server and storage medium for ordering multimedia data | |
| CN114020999A (en) | Community structure detection method and system for movie social network | |
| CN115618101A (en) | Streaming media content recommendation method and device based on negative feedback and electronic equipment | |
| Fazelnia et al. | Variational user modeling with slow and fast features | |
| CN114637909A (en) | Film recommendation system and method based on improved deep structured semantic model | |
| CN112231579B (en) | Social video recommendation system and method based on implicit community discovery | |
| CN113468413B (en) | Multi-user sharing-oriented multimedia network video recommendation method | |
| CN114117233A (en) | Conversation news recommendation method and system based on user implicit feedback | |
| CN117056609A (en) | Session recommendation method based on multi-layer aggregation enhanced contrast learning | |
| CN110769288A (en) | Video cold start recommendation method and system | |
| CN116662656A (en) | Movie recommendation method based on collaborative enhancement and graph annotation intention neural network | |
| Mirhasani et al. | Alleviation of cold start in movie recommendation systems using sentiment analysis of multi-modal social networks | |
| CN114780841A (en) | KPHAN-based sequence recommendation method | |
| CN114491272A (en) | Multimedia content recommendation method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |