CN114936317A - Marriage dating recommendation method and device based on image content analysis - Google Patents
Marriage dating recommendation method and device based on image content analysis Download PDFInfo
- Publication number
- CN114936317A CN114936317A CN202210465285.0A CN202210465285A CN114936317A CN 114936317 A CN114936317 A CN 114936317A CN 202210465285 A CN202210465285 A CN 202210465285A CN 114936317 A CN114936317 A CN 114936317A
- Authority
- CN
- China
- Prior art keywords
- user
- model
- face
- recall
- marriage
- 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 52
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000001514 detection method Methods 0.000 claims abstract description 16
- 230000006399 behavior Effects 0.000 claims description 26
- 230000008569 process Effects 0.000 claims description 12
- 238000001914 filtration Methods 0.000 claims description 9
- 238000013500 data storage Methods 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000004044 response Effects 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 claims description 3
- 230000008878 coupling Effects 0.000 claims description 2
- 238000010168 coupling process Methods 0.000 claims description 2
- 238000005859 coupling reaction Methods 0.000 claims description 2
- 230000001815 facial effect Effects 0.000 claims 1
- 239000011435 rock Substances 0.000 claims 1
- 238000012546 transfer Methods 0.000 description 22
- 238000009434 installation Methods 0.000 description 15
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 12
- 238000010586 diagram Methods 0.000 description 7
- 230000007246 mechanism Effects 0.000 description 7
- 238000001816 cooling Methods 0.000 description 5
- 238000002955 isolation Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000010354 integration Effects 0.000 description 2
- 230000003698 anagen phase Effects 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 229910001220 stainless steel Inorganic materials 0.000 description 1
- 239000010935 stainless steel Substances 0.000 description 1
- 230000000007 visual effect 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
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/16—Constructional details or arrangements
- G06F1/20—Cooling means
-
- 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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Human Computer Interaction (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Computational Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a marriage and love dating recommendation method based on image content analysis. The marriage dating recommendation method based on image content analysis comprises the following steps: s1, training a face recognition model and a face detection model by using the public data set, taking the models as extractors of image characteristics, and extracting the image characteristics of a user photo in an off-line manner; and S2, according to the user behavior data, training ALS in an off-line mode to serve as a recall model, and generating a user list to be recommended for each user. According to the marriage dating recommendation method based on image content analysis, provided by the invention, the image characteristics are added, the preference of the user on the image content can be effectively captured, the dimensionality of the portrait of the user is increased, and the accuracy degree of the model on the recommended content is ensured, so that the spouse selecting object matched with the model can be efficiently and accurately recommended for the user, and the spouse selecting efficiency of the user is improved.
Description
Technical Field
The invention relates to the technical field of marriage dating recommendation, in particular to a method and a device for marriage dating recommendation based on image content analysis.
Background
Along with the continuous development of human life, marriage is taken as a life event which is required to be experienced by people, and is generated by love between men and women, along with the requirement of work, the opportunity of traveling outside is reduced, and the opposite friends who contact with the marriage are difficult to be accepted by both parties, so that the number of love and love is sharply reduced, and the way of making friends on network becomes the development trend of novel friends in the future.
The marriage and love dating website provides a platform for meeting and acquaintance of single men and women, but enables the single men and women to be unavailable in massive information and various choices, and in order to accurately find a proper couple selecting object for a user, the recommendation system takes effect, and the marriage and love dating recommendation system is based on the couple selecting requirements of internet users and utilizes a recommendation algorithm to filter and screen, so that the couple selecting object which is possibly interested in the user is pushed to the user.
The traditional marriage and love dating recommendation mainly matches users based on information such as user attributes and user behaviors, and cannot effectively integrate multimodal data such as user pictures and characters for association and recommendation, and in a scene of the marriage and love dating, visual elements such as a growth phase and appearances often have higher priority in an idol selection process, so that image content analysis is added in the marriage and love dating recommendation, the current requirements of the users are judged according to the similarity of the image contents, and the method is an important way for improving the matching degree of the marriage and love users.
Therefore, it is necessary to provide a dating recommendation method and apparatus based on image content analysis to solve the above technical problems.
Disclosure of Invention
The invention provides a marriage and dating recommendation method based on image content analysis, which solves the problem that the integration degree of online marriage and dating recommendation user information needs to be improved.
In order to solve the technical problems, the invention provides a marriage and love dating recommendation method based on image content analysis, which comprises the following steps:
s1, training a face recognition model and a face detection model by using the public data set, taking the face recognition model and the face detection model as an extractor of image features, and extracting the image features of the user photo offline;
s2, according to the user behavior data, taking offline training ALS as a recall model, and generating a user list to be recommended for each user;
s3, recommending similar face users for each user according to the favorite heterologism photo of the user and combining the face image characteristics of the photo to expand a user list to be recommended, comprehensively considering face similarity and ALS model scoring conditions, and reordering recalls;
s4, acquiring basic attributes, puppet selection conditions and image characteristic data of a user, clicking the data by the user, and training a Wide & Deep model for online sequencing;
and S5, providing user recommendation service by means of a Web service technology, wherein the service can realize the processes of ALS plus image feature recall, W & D model fine-ranking and business rule filtering, and realize personalized accurate recommendation for each user.
Preferably, the extractor of the image features in step S1 includes a face detection model and a face feature extraction model, where the face detection model uses a scrfd model, and the model obtains an optimal network structure by a network search method on the premise that a calculated amount is fixed, and guarantees balance of partial structures of backbone, tack, and head, and improves accuracy of model detection by strategies of training sample balance, ATSS, and DIOU, so as to obtain a training model.
Preferably, the face recognition model adopts an insight surface structure, and the insight surface network improves the separability between classes in the spatial distribution of the face features by adding a margin to the loss function, and strengthens the difference between the class internal tightness and the class, so as to improve the face recognition precision, and for the model to have better real-time performance, the backbone network uses a lightweight mobilenetv2 structure.
Preferably, the training model is used for extracting the face features of the picture uploaded by the user, and the result is stored in the hive table and is convenient to read during recall.
Preferably, the recall user in step S3 adopts a similar face recall strategy to recommend a photo similar to the user who likes it, and performs PQ encoding on the face features.
Preferably, the PQ coding method is to divide 512-dimensional face features into N-dimensional subspaces, perform M-class clustering by using a Kmeans algorithm in each subspace, use a clustered class ID as a PQ code, acquire a feature code near the PQ code because the cluster center is distributed in the feature space, perform matching according to adjacent feature code values, and greatly improve the global search speed.
Preferably, the PQ codes of the photos of the user and the adjacent PQ codes are written into a hive table respectively in the process of off-line searching, the adjacent PQ codes of the photos that the user likes and the PQ codes of the photos to be searched are used as matching rules, a fast off-line global searching function is realized by means of the computation capability of spark, and the recall result is roughly ranked by combining the scoring of ALS and the similarity of faces in step S2.
Preferably, in the step S4, basic attributes of the user are obtained, where the basic attributes include, but are not limited to, gender, age, height, constellation, hobby, income, province of long living, county and city, even-preference conditions, and color data, the user can slide left and right on a card recommendation page, a wide & deep model is trained, an Accuracy, call, F1-score, AUC indexes, and giuc indexes are integrated to obtain an optimal offline model under current conditions, in order to improve the response speed of a recommendation service end, model parameters of offline training and encoding results of user features are cached respectively, and an online reasoning part uses a wide & deep frame self-built based on a pytorh to realize an ms-level reference.
Preferably, the ALS recall pool performs some condition filtering operations in the early stage, results are directly taken for use, the data structure is simple, and Redis is used for caching; for new users, the group does not generate behaviors, so that no corresponding recall result exists in an ALS recall pool, Mongodb is used for caching active users, some key matching information is added, matching information of the users is used as a condition in the recall process to search target users meeting the condition in the cache pool, the W & D model is trained in step S4 to enter an online W & D inference logic to obtain recall results as candidates, and the recommendation is preferentially carried out according to scoring results finally output by the W & D model.
The invention also provides a marriage and love dating recommendation device based on image content analysis, which comprises:
the system comprises a user face image characteristic database, a user attribute and behavior data database, a user recall list database and a user recommendation web server;
the user image feature database comprises a user id, a picture id, a face feature code and a hive table, and data of all user photos are stored in the hive table after offline operation;
the user attribute and behavior data database comprises basic information of user id, gender, age and height and puppet selection conditions, and behavior data clicked and browsed by a user are stored in a hive table;
the user recall list database comprises a user id, a user id to be recommended, a redis database and a redis database for data storage;
the user recommendation web server comprises a module for acquiring a user to be recommended, a module for acquiring user basic attributes and puppet selection condition characteristics, a module for acquiring image characteristics, a module for reasoning Wide & Deep models and a module for filtering business rules.
Compared with the related art, the marriage dating recommendation method based on image content analysis provided by the invention has the following beneficial effects:
the invention provides a marriage and love dating recommendation method based on image content analysis, which is added with image features, can effectively capture the preference of a user to image content, increases the dimensionality of a portrait of the user, and ensures the accuracy degree of a model to recommended content, thereby efficiently and accurately recommending a spouse selecting object matched with the user for the user, and improving the spouse selecting efficiency of the user.
Drawings
Fig. 1 is a flowchart of a method for recommending dating based on image content analysis according to the present invention;
FIG. 2 is a flowchart of an image search of a dating recommendation method based on image content analysis according to the present invention;
FIG. 3 is a block diagram of a web interface portion of a dating recommendation method based on image content analysis according to the present invention;
fig. 4 is a schematic structural diagram of a dating recommendation device for marriage dating based on image content analysis according to the present invention;
fig. 5 is a schematic structural diagram of a server installation device of the dating recommendation method based on image content analysis according to the present invention;
FIG. 6 is a schematic structural view of the heat exchange mechanism part shown in FIG. 5;
fig. 7 is a schematic view of the structure of the mount section shown in fig. 5.
The reference numbers in the figures:
1. a mounting frame;
2. a heat exchange box 201, an adjusting hole 210, a circulation cavity 220 and a heat exchange cavity;
3. a telescoping member;
4. a lifting frame 41 and an installation platform;
5. an isolation net;
6. an adjusting plate;
7. mounting rack 71, ice bag;
8. the heat exchange device comprises a heat exchange mechanism 81, a water pump 82, a heat exchange pipe 83 and a heat exchange rod.
Detailed Description
The invention is further described with reference to the following figures and embodiments.
Please refer to fig. 1, fig. 2, fig. 3, fig. 4, fig. 5, fig. 6, and fig. 7 in combination, wherein fig. 1 is a flowchart of a method for dating recommendation based on image content analysis according to the present invention; FIG. 2 is a flowchart of an image search of a dating recommendation method based on image content analysis according to the present invention; FIG. 3 is a block diagram of a web interface portion of a dating recommendation method based on image content analysis according to the present invention; fig. 4 is a schematic structural diagram of a dating recommendation device for marriage dating based on image content analysis according to the present invention; fig. 5 is a schematic structural diagram of a server installation device of the marriage and dating recommendation method based on image content analysis according to the present invention; FIG. 6 is a schematic structural diagram of a heat exchange mechanism part shown in FIG. 5; fig. 7 is a schematic view of the structure of the mounting bracket portion shown in fig. 5.
A marriage and love dating recommendation method based on image content analysis comprises the following steps:
s1, training a face recognition model and a face detection model by using the public data set, taking the face recognition model and the face detection model as an extractor of image features, and extracting the image features of the user photo offline;
the extractor of the image features in the step S1 includes a face detection model and a face feature extraction model, where the face detection model uses a scrfd model, and the model obtains an optimal network structure by a network search method on the premise of fixed computation, and ensures that the backbone, tack, and head partial structures are balanced, and improves the accuracy of model detection by training sample balance, ATSS, and DIOU strategies to obtain a training model;
the human face recognition model adopts an insight face structure, and the insight face network improves the separability among classes in the spatial distribution of human face features by adding a margin to a loss function, and strengthens the difference between class internal tightness and class difference, so that the human face recognition precision is improved, and for better real-time performance of the model, a light-weight mobilenetv2 structure is used as a main network;
using the training model to extract the face features of the picture uploaded by the user, and storing the result in a hive table for convenient reading during recall;
s2, according to the user behavior data, training ALS in an off-line mode to serve as a recall model, and generating a user list to be recommended for each user;
collecting behavior data of the user, wherein a large and representative amount of behavior data generated by the user on the APP is selected, for example: the method comprises the steps that a User slides and clicks data left and right on a recommended page of a home page card, the User sends and reads credit data, the User browses data of other people's homepages, the User browses, approves and reviews behavior data of a page dynamically, different behavior scores are manually set for each behavior according to behavior categories of the User and the conformity degree of a final target, meanwhile, in order to reflect the difference of the same behavior, time attenuation factors are constructed, the behavior scores are weighted by using the time attenuation factors when the behaviors are scored, finally, the excellent processing performance of spark on big data is utilized, all behavior data are combined to obtain a final User-Iterm Rating Matrix, then, an ALS model is used for explicit feedback training, the model obtained by training is utilized for carrying out personalized recommendation on each User, and a personalized recall pool of the User is obtained;
s3, recommending similar face users for each user according to the favorite heterographs of the users and combining the face image characteristics of the photos to expand a user list to be recommended, comprehensively considering face similarity and ALS model scoring conditions, and using recalls for reordering;
the recalling user in the step S3 adopts a similar face recalling strategy to recommend photos similar to the user liked by the user for the user, and PQ coding is carried out on the face features;
the PQ coding method comprises the steps of dividing 512-dimensional face features into N-dimensional subspaces, clustering M categories in each subspace by using a Kmeans algorithm, taking the clustered category IDs as PQ codes, acquiring feature codes near the PQ codes due to the fact that clustering centers are distributed in the feature spaces, matching according to adjacent feature code values, and greatly improving the overall searching speed;
in the off-line searching process, the PQ codes of the user photos and the adjacent PQ codes are respectively written into a hive table, the adjacent PQ codes of the photos which the user likes and the PQ codes of the photos to be searched are used as matching rules, the fast off-line global searching function is realized by means of the computing power of spark, and the recalling result is roughly ranked by combining the scoring of ALS and the similarity of faces in the step S2;
s4, acquiring basic attributes, puppet selection conditions and image characteristic data of a user, clicking the data by the user, and training a Wide & Deep model for online sequencing;
the basic attributes of the user are obtained in the step S4, the basic attributes include but are not limited to sex, age, height, constellation, hobbies, income, province of long-living work, county and city, coupling conditions and color value data for integration, the user can slide left and right on a card recommendation page, a wide & deep model is trained, an optimal offline model under the current condition is obtained by integrating Accuracy, Recall, F1-score, AUC indexes and giuc indexes, model parameters of offline training and encoding results of user characteristics are cached respectively in order to improve the response speed of a recommendation service end, and an online reasoning part uses a wide & deep frame self-built based on pytorh to realize an ms-level reference;
s5, providing user recommendation service by means of Web service technology, wherein the service can realize the processes of ALS plus image feature recall, W & D model fine-ranking and business rule filtering, and realize personalized accurate recommendation for each user;
the ALS recall pool performs some condition filtering operations in the early stage, results are directly taken for use, the data structure is simple, and Redis is used for caching; for new users, the group does not generate behaviors, so that no corresponding recall result exists in an ALS recall pool, Mongodb is used for caching active users, some key matching information is added, matching information of the users is used as a condition in the recall process to search target users meeting the condition in the cache pool, the W & D model is trained in step S4 to enter an online W & D inference logic to obtain recall results as candidates, and the recommendation is preferentially carried out according to scoring results finally output by the W & D model.
The invention also provides a marriage and love dating recommendation device based on image content analysis, which comprises:
the system comprises a user face image feature database, a user attribute and behavior data database, a user recall list database and a user recommendation web server;
the user image feature database comprises a user id, a picture id, a face feature code and a hive table, and data of all user photos are stored in the hive table after offline operation;
the user attribute and behavior data database comprises basic information of user id, sex, age and height and puppet selection conditions, and behavior data clicked and browsed by a user are stored in the hive table;
the user recall list database comprises a user id, a user id to be recommended, a redis database and a redis database for data storage;
the user recommendation web server comprises a module for acquiring a user to be recommended, a module for acquiring user basic attributes and puppet selection condition characteristics, a module for acquiring image characteristics, a module for reasoning Wide & Deep models and a module for filtering business rules.
The face similarity is trained and analyzed through the image model, and the personalized information of the users is fully considered, so that the matching degree between the users is effectively improved, the problem that image content information is lacked in a traditional recommendation algorithm is solved, the efficiency of finding the best partner is improved for the users, and the user experience and the user activity are improved.
Compared with the related art, the marriage dating recommendation method based on image content analysis provided by the invention has the following beneficial effects:
the method adds the image characteristics, can effectively capture the preference of the user to the image content, increases the dimensionality of the portrait of the user, and ensures the accuracy degree of the model to the recommended content, thereby efficiently and accurately recommending the puppet selection object matched with the puppet selection object for the user, and improving the puppet selection efficiency of the user.
The invention provides a marriage and love dating recommendation device based on image content analysis, which needs to use a server installation device in the installation and use processes, wherein the server installation device comprises:
the mounting frame 1 is fixedly connected with the heat exchange box 2, and the heat exchange box 2 is provided with an adjusting hole 201;
the telescopic piece 3 is fixedly arranged on the mounting frame 1;
the lifting frame 4 is fixedly arranged at the shaft end of the telescopic part 3, the lifting frame 4 is of an L-shaped structure, and the mounting table 41 is arranged on the lifting frame 4;
the isolation net 5 is fixedly installed in the heat exchange box 2, and the isolation net 5 divides the heat exchange box 2 into a circulation cavity 210 and a heat exchange cavity 220;
the adjusting plate 6 is movably arranged on the heat exchange box 2;
the mounting frame 7 is fixedly arranged on the adjusting plate 6, an ice bag 71 is arranged on the mounting frame 7, and the mounting frame 7 is inserted into the heat exchange cavity 220 through the adjusting hole 201;
The extensible member 3 adopts electric telescopic handle, connects external power during the use, provides stable power source for the lift adjustment of crane 4.
The inside injection of heat transfer case 2 has the heat transfer water source, and heat transfer case 2 internally mounted has water pump 81, connects external power during the use of water pump 81, and the water source after for the inside heat transfer of circulation chamber 210 is drained, and the heat transfer water source carries out the heat transfer through heat exchange tube 82 and heat transfer pole 83 and the inside air of mounting bracket 1, provides reliable and stable cooling source for the space of server installation, keeps stable heat transfer cooling and operation under the server closed condition.
The isolation net 5 adopts the stainless steel net, provides spacingly for the installation of mounting bracket 7, and mounting bracket 7 makes things convenient for the installation of ice bag 71, can cool off the heat transfer to the inside water source in heat transfer chamber 220 after the installation of ice bag 71, and the in-range of water source after the heat transfer can be retransmitted to ice bag 71 to the heat exchange tube 82 output, realizes the circulative cooling of water source after the heat transfer.
The adjusting plate 6 conveniently drives the ice bag 71 to lift and adjust through the mounting frame 7, and the ice bag 71 is convenient to mount and dismount.
The server installation equipment provided by the invention has the beneficial effects that:
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A marriage and love dating recommendation method based on image content analysis is characterized by comprising the following steps:
s1, training a face recognition model and a face detection model by using the public data set, taking the face recognition model and the face detection model as an extractor of image features, and extracting the image features of the user photo offline;
s2, according to the user behavior data, training ALS in an off-line mode to serve as a recall model, and generating a user list to be recommended for each user;
s3, recommending similar face users for each user according to the favorite heterographs of the users and combining the face image characteristics of the photos to expand a user list to be recommended, comprehensively considering face similarity and ALS model scoring conditions, and using recalls for reordering;
s4, acquiring basic attributes, puppet selection conditions and image characteristic data of a user, clicking the data by the user, and training a Wide & Deep model for online sequencing;
and S5, providing user recommendation service by means of a Web service technology, wherein the service can realize the processes of ALS and image feature recall, W & D model fine-ranking and business rule filtering, and realize personalized accurate recommendation for each user.
2. The image content analysis-based marriage dating recommendation method according to claim 1, wherein the image feature extractor in step S1 comprises a face detection model and a face feature extraction model, wherein the face detection model uses a scrfd model, the model uses a network search method to obtain an optimal network structure under the premise of fixed computation amount, and to ensure that the backbone, rock, and head partial structures are balanced, and the training model is obtained by training the strategies of sample balancing, ATSS, and DIOU to improve the accuracy of model detection.
3. The image content analysis-based dating recommendation method according to claim 2, wherein the face recognition model adopts an insight face structure, and the insight face network increases inter-class separability in spatial distribution of face features by adding margin to a loss function, and enhances intra-class compactness and inter-class differences, thereby improving face recognition accuracy, and for better real-time performance of the model, the backbone network uses a lightweight mobilenetv2 structure.
4. A marriage and love dating recommendation method according to claim 3, wherein said training model is used to extract the facial features of the user's uploaded photos and store the results in the hive table for easy recall.
5. A marriage and love dating recommendation method according to claim 4, wherein said step S3 is implemented by using a similar face recall strategy to recall users, and recommending photos similar to the users who like the users for the users, and PQ coding the face features.
6. The image content analysis-based dating recommendation method according to claim 5, wherein the PQ coding method is characterized in that 512-dimensional face features are divided into N-dimensional subspaces, each subspace is clustered by M classes using a Kmeans algorithm, the clustered class IDs are used as PQ codes, and since the cluster centers are distributed in the feature space, feature codes near the PQ codes are obtained, and matching is performed according to adjacent feature code values, so that the global search speed is greatly increased.
7. A marriage and love dating recommendation method according to claim 6, wherein the PQ codes of the photos of the user and the PQ codes adjacent to the photos are written into the hive table during the off-line search process, the PQ codes adjacent to the photos liked by the user and the PQ codes of the photos to be searched are used as matching rules, and the fast off-line global search function is realized by means of the computation capability of spark, and the recall result is roughly ranked by combining the score of ALS and the face similarity in step S2.
8. The method of claim 7, wherein basic attributes of the user are obtained in step S4, the basic attributes include but are not limited to gender, age, height, constellation, hobbies, income, province of long-lived work, county and city, coupling conditions, and color data, the user can slide left and right on a card recommendation page, a wide & deep model is trained, an Accuracy, Recall, F1-score, AUC indexes, and giuc indexes are integrated to obtain an optimal offline model under current conditions, model parameters of our online training and encoding results of user features are respectively cached in order to improve response speed of a recommendation service end, and an online reasoning part realizes an ms-level reference using a wide & deep frame self-built based on pytorh.
9. The image content analysis-based marriage dating recommendation method according to claim 8, wherein the ALS recall pool is subjected to some conditional filtering operations in the early stage, and the results are directly taken for use, and the data structure is simple and is cached by using Redis; for new users, the group does not generate behaviors, so that a corresponding recall result does not exist in an ALS recall pool, Mongoldb is used for caching some active users, some key matching information is added, matching information of the users is used as a condition in the recall process to search target users meeting the condition in the cache pool, the W & D model is trained in step S4 to use the obtained recall result as a candidate to enter online W & D reasoning logic, and recommendation is preferentially carried out according to a scoring result finally output by the W & D model.
10. A marriage and love dating recommendation apparatus based on image content analysis, comprising the marriage and love dating recommendation method based on image content analysis according to any one of claims 1 to 9, comprising:
the system comprises a user face image feature database, a user attribute and behavior data database, a user recall list database and a user recommendation web server;
the user image feature database comprises a user id, a picture id, a face feature code and a hive table, and data of all user photos are stored in the hive table after offline operation;
the user attribute and behavior data database comprises basic information of user id, gender, age and height, and behavior data of puppet selection condition, user clicking and browsing, and is stored in a hive table;
the user recall list database comprises a user id, a user id to be recommended, a redis database and a redis database for data storage;
the user recommendation web server comprises a module for acquiring a user to be recommended, a module for acquiring user basic attributes and puppet selection condition characteristics, a module for acquiring image characteristics, a module for Wide & Deep model reasoning and a module for filtering service rules.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210465285.0A CN114936317A (en) | 2022-04-29 | 2022-04-29 | Marriage dating recommendation method and device based on image content analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210465285.0A CN114936317A (en) | 2022-04-29 | 2022-04-29 | Marriage dating recommendation method and device based on image content analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114936317A true CN114936317A (en) | 2022-08-23 |
Family
ID=82862464
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210465285.0A Pending CN114936317A (en) | 2022-04-29 | 2022-04-29 | Marriage dating recommendation method and device based on image content analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114936317A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103678394A (en) * | 2012-09-21 | 2014-03-26 | 孟露芳 | Image matching degree based marriage dating recommendation method and system |
| CN107291841A (en) * | 2017-06-01 | 2017-10-24 | 广州衡昊数据科技有限公司 | A kind of method and system based on position and the social target of user's portrait intelligent Matching |
| US20200050627A1 (en) * | 2017-03-29 | 2020-02-13 | Panasonic Corporation | Communication device, communication system, and content collection method |
| WO2020141799A2 (en) * | 2019-01-03 | 2020-07-09 | 오드컨셉 주식회사 | Method for providing fashion item recommendation service to user by using calendar data |
| CN113641904A (en) * | 2021-08-16 | 2021-11-12 | 上海花千树信息科技有限公司 | Method and device for recommending target customers for stores under Internet marriage line |
-
2022
- 2022-04-29 CN CN202210465285.0A patent/CN114936317A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103678394A (en) * | 2012-09-21 | 2014-03-26 | 孟露芳 | Image matching degree based marriage dating recommendation method and system |
| US20200050627A1 (en) * | 2017-03-29 | 2020-02-13 | Panasonic Corporation | Communication device, communication system, and content collection method |
| CN107291841A (en) * | 2017-06-01 | 2017-10-24 | 广州衡昊数据科技有限公司 | A kind of method and system based on position and the social target of user's portrait intelligent Matching |
| WO2020141799A2 (en) * | 2019-01-03 | 2020-07-09 | 오드컨셉 주식회사 | Method for providing fashion item recommendation service to user by using calendar data |
| CN113641904A (en) * | 2021-08-16 | 2021-11-12 | 上海花千树信息科技有限公司 | Method and device for recommending target customers for stores under Internet marriage line |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110232152B (en) | Content recommendation method, device, server and storage medium | |
| CN111177446B (en) | Method for searching footprint image | |
| US7853582B2 (en) | Method and system for providing information services related to multimodal inputs | |
| US20080052312A1 (en) | Image-Based Face Search | |
| CN102549603A (en) | Relevance-based image selection | |
| Chen et al. | Research on personalized recommendation hybrid algorithm for interactive experience equipment | |
| US8370323B2 (en) | Providing information services related to multimodal inputs | |
| CN114461839B (en) | Multi-mode pre-training-based similar picture retrieval method and device and electronic equipment | |
| KR20200083159A (en) | Method and system for searching picture on user terminal | |
| CN113596601A (en) | Video picture positioning method, related device, equipment and storage medium | |
| CN111949806A (en) | Cross-media retrieval method based on Resnet-Bert network model | |
| CN102722528A (en) | Real-time mass image retrieval method based on mobile devices | |
| CN111538818A (en) | Data query method and device, electronic equipment and storage medium | |
| CN112347339A (en) | Search result processing method and device | |
| CN107391599B (en) | Image retrieval method based on style characteristics | |
| Li et al. | Personalized query auto-completion for large-scale POI search at Baidu Maps | |
| CN109635277A (en) | A kind of method and relevant apparatus obtaining entity information | |
| Chandrakala et al. | Application of artificial bee colony optimization algorithm for image classification using color and texture feature similarity fusion | |
| CN114936317A (en) | Marriage dating recommendation method and device based on image content analysis | |
| CN117235362A (en) | Analysis system based on big data of wisdom text travel | |
| CN110110126A (en) | Inquire the method, apparatus and server of the face-image of personage | |
| CN111831847B (en) | Similar picture set recommendation method and system | |
| US11727051B2 (en) | Personalized image recommendations for areas of interest | |
| KR100890253B1 (en) | Method For Acquiring And Providing Knowledge Using Human Agents On Network For Oncoming Generation And The System Therefor | |
| CN116226320A (en) | Method and device for predicting context information, electronic equipment and storage medium |
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 |