CN109584003A - Intelligent recommendation system - Google Patents
Intelligent recommendation system Download PDFInfo
- Publication number
- CN109584003A CN109584003A CN201811417462.8A CN201811417462A CN109584003A CN 109584003 A CN109584003 A CN 109584003A CN 201811417462 A CN201811417462 A CN 201811417462A CN 109584003 A CN109584003 A CN 109584003A
- Authority
- CN
- China
- Prior art keywords
- user
- product
- recommendation
- behavior
- module
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Abstract
A kind of intelligent recommendation system, the system include modules A, module B and module C.Modules A converts behavioural characteristic after extracting from user behavior data library to user behavior, and the feature vector of user is generated in conjunction with user attribute database;Module B generates feature-article associated recommendation matrix from multiple correlation tables, and the feature vector of user is converted into initial recommendation item lists by feature-article associated recommendation matrix;Module C, be responsible for initial recommendation list is filtered, ranking processing, to generate consequently recommended result.
Description
Technical field
The invention belongs to Internet technical field, in particular to a kind of intelligence for user's recommendation service or product
Recommender system.
Background technique
Intelligent recommendation system, also referred to as " recommender system ", since the nineties in last century is born, always in internet
Various fields play central role.Such as: the e-commerce system of Amazon, the video recommendation system of youku.com, Pandora sound
Happy recommender system etc..The problem of the difficult always internet development of the selection of consumer, for example, thousands of portions on large-scale website
Film can see but be unknown to which portion seen by point;Millions of first songs can play in real time, but can not find and want to listen.Letter
While breath technology brings information season to enrich, the selection difficulty disease of user is also brought.And best selection, it does not need exactly
Selection.Develop by more than ten years technology, nowadays problem to be solved gradually emerges.It is right in the environment of information overload
The offer of information consumption demand possible solution that here it is recommender systems.
The high speed development of mobile Internet in recent years, mobile personal terminal device are more and more abundant.Often mobile device body
Product is smaller, and page carrying information content is limited, this means that the information showed needs more accurately meet each user's
Humanized demand, and the diversity of the fragmentation of media culture, also bring demand of the people to information more and more personalized.These
Factors drive the universalness and deepization of personalized and recommender system application.
Traditional way of recommendation needs to build the management backstage of an offer human configuration, and the product of recommendation in need passes through
Artificial screening delineation, sequence, then the rule of the meeting human configuration of the script by automating pull product data, client from lane database
Termination is shown that user has more the needs of product shown is selected and meets oneself by product data foreground.Such process task
Process is longer, largely needs by artificial experience intervention, and artificial getting sth into one's head can generate error and inefficient situation, to give
The accuracy of recommender system impacts.
Since efficient data retrieval and service efficiency are always the core of recommender system, application scenarios is being recommended to reach into
Thousand pages up to a hundred, carrying million data demands up to a hundred when calculating in real time, depend merely on artificial experience and go substitution operation, be to be difficult to imagine
's.Moreover there are human errors in calculating process, inevitably will appear recommendation not in time, there is phenomena such as accidentally recommending.
Summary of the invention
The embodiment of the invention provides a kind of intelligent recommendation system, for by inline system to user's recommendation service or
Product, these products or service content are extensive, including travelling products, films and television programs and extensive stock.
One of the embodiment of the present invention, a kind of intelligent recommendation system, the system include modules A, module B and module C.
Modules A converts behavioural characteristic after extracting from user behavior data library to user behavior, in conjunction with user property
The feature vector of database generation user;
Module B generates feature-article associated recommendation matrix from multiple correlation tables, and the feature vector of user passes through feature-object
Product associated recommendation matrix is converted into initial recommendation item lists;
Module C, be responsible for initial recommendation list is filtered, ranking processing, to generate consequently recommended result.
The recommender system includes candidate products set, and feature-article which is supplied to module B is related
Recommend matrix reference, while being also provided in the filter process of module C and reference is provided.
Also ranking processing is supplied to by user behavior feedback and goods attribute in module C to be referred to.
Intelligent recommendation system of the invention is by big data processing capacity, it is possible to provide real-time data operation and retrieval energy
Power greatly improves the performance that data use.In addition, recommender system is applied in multiple complex pages, it can be real by algorithm
The personalized service in existing " thousand people, thousand face " provides strength guarantee for accuracy, the diversity of recommendation.
As various physical entity Intellectualized Tendencies are more and more obvious, the birth of intelligent recommendation system is to pure digi-tal generation
It explores to boundary, the premise that recommender system eventually can generate those between user and product connection is found out, and allows product
It is really connected with user.It is no longer that people looks for product with the arrival that all things on earth interconnects, product looks for the trend of people head-on to come up, and
It can be further strengthened.Intelligent recommendation is exactly to be born in the environment of such digitlization, and constantly have found that it is likely that meeting and people
The another kind of product node connected provides personalized recommendation service, meets the needs of different, and perhaps this is only intelligence and pushes away
System is recommended, realizes that people is the ultimate significance of interconnection.
Detailed description of the invention
The following detailed description is read with reference to the accompanying drawings, above-mentioned and other mesh of exemplary embodiment of the invention
, feature and advantage will become prone to understand.In the accompanying drawings, if showing by way of example rather than limitation of the invention
Dry embodiment, in which:
Intelligent recommendation systematic difference logical schematic in Fig. 1 embodiment of the present invention.
Intelligent recommendation system engine structural schematic diagram in Fig. 2 embodiment of the present invention.
Collaborative filtering example in Fig. 3 embodiment of the present invention.
Specific embodiment
There are a UI system, UI, that is, User Interface at the end PC of website access or the mobile end APP
The abbreviation of (user interface).UI system is responsible for user's displayed web page and and user's interaction.This UI system can pass through day aspiration
Into the log of user behavior, log is potentially stored in memory cache the various behavior record united by user on UI,
Storage is also possible in the database, when recommender system is needed with these user behaviors logs, from memory cache or database
It takes, so as to further analyze user behavior, generates the recommendation list that portion tries to figure out user intent, finally return again
To UI system demonstration on the interface of PC or APP.
As shown in Figure 1, intelligent recommendation system needs play a role other than the recommendation function of recommender system, boundary is also relied on
Face shows UI and user behavior data.Showing interface UI effect is mainly shown product and product attribute by data transfer mode
At the moment to user, the miscellaneous recommendation position such as in the terminals such as the station PC, the station H5, wechat, APP.In addition to product itself, in order to
Increase the attraction of user, many times can also the product attributes such as affix product title, thumbnail, product introduction, intelligently push away
The system of recommending can according to these product attributes of the target adjustment of recommendation, such as: promote user to place an order, increase preferential promotion label, increasing
Add tag along sort, for the product introduction information coloring recommended originally.Meanwhile user can be appreciated that recommender system recommend product when,
Oneself interbehavior can be generated on website, for example, click, browse, place an order the product, these are referred to as the number of user behavior
According to the data target of frequent user behavior such as following table.
At this point, User action log storage system is started to work, user behavior data is collected and stored, intelligence is finally provided
Recommended engine framework statistics, uses modeling, finally exports new proposed algorithm, is presented to product again according to algorithmic rule
On each interface UI of PC or APP, to realize the effect of personalized recommendation, to promote more good user service.If table 1 is user's row
For the example of data.
Table 1
Behavior | User type | Scale | Real time access |
Browse webpage | Registration/anonymity | Greatly | × |
Buy product | Registration | In | √ |
Collect product | Registration | In | √ |
Comment on product | Registration | It is small | √ |
It scores to product | Registration | It is small | √ |
Search for product | Registration/anonymity | Greatly | × |
Click search result | Registration/anonymity | Greatly | × |
Share product | Registration | It is small | √ |
The core of intelligent recommendation system is data collection and storage, wherein user behavior data is introduced, Yi Jiyong
How family behavioral data acts on.Below under user behavior data introduction, this partial data how to obtain and data
Characteristic which has, illustrate in terms of three respectively: 1) Production conditions, 2) scale, 3) access mode illustrates in intelligent recommendation
How to obtain and use in system.As can be seen from Table 1:
1) in terms of Production conditions, some behaviors are that only registration user could generate;Some behaviors are that all users may be used
With generation.Such as: browsing webpage, clicks search product at search product
2) from scale, browsing webpage, the scale of search record are all very big, because all users of this behavior can produce
It is raw, and averagely everyone can generate these behaviors.Purchase, collection behavior scale are medium, because only that registration user's ability
This behavior is generated, but buying behavior is the main behavior of website again, so they are larger for comment, but phase
Scale is much smaller for web page browsing behavior.Last remaining behavior is that the sub-fraction talent registered in user has
, so scale will not be very big.
3) from the angle of real time access, the behaviors such as buying, collect, commenting on, scoring, sharing all is to need real time access
, as long as need to be embodied on interface because user has these behaviors, such as after user has purchased product, of user
People buys the product that should just show user's purchase in list immediately.And some behaviors, such as behavior and the search row of browsing webpage
Not need to obtain in real time.
In intelligent recommendation system of the invention, the data for needing real time access are stored in database and caching, greatly
Scale accesses data in non real-time and is stored in distributed file system (HDFS).Can data real time access depends on take
To the new behavior of user, the new behavior of user is only taken, recommender system could be applicable in the current demand of user in real time, to use
Family carry out real-time recommendation, therefore data can real time access it is increasingly important in intelligent recommendation system.
According to one or more embodiment, the exchange architecture of intelligent recommendation system of the invention needs to consider two o'clock: 1)
Real-time 2) accuracy.
Real-time allows for real-time response user request, and one has the system of real-time response user request, not only exists
User side can capture more rapidly intention, and in Products Show requiring the product of real time price to be pushed to user.Such as trip
Trip has the air ticket and train ticket of many real time prices on website, can be to using the characteristics of real-time when recommending for these products work
The pricing information of family offer What You See Is What You Get.Currently, most of recommender system existing in the market, due to by business or technology
It restrains and delay or the recommendation function of T+1 is often provided, such system is less to news category and the recommendation of tourism industry air ticket
Friendly, this proposes the real-time of the recommender system framework of report it is possible to prevente effectively from problems.
The considerations of accuracy, is divided to two aspects, is on the one hand that Products Show is correct, it is a large amount of expired on tour site or
Person's product that fails needs to filter when recommending, and what is showed to user is the product that can actually sell, accomplish user see i.e. obtained by;Separately
On the one hand, recommend him interested according to user interest, i.e., " needed for thinking that user thinks, recommends ".Both comprehensive to data
It records, parse, use in algorithm, foreground feedback user behavioral data forms closed loop.Based on above-mentioned two o'clock recommender system engine
Framework is designed into 3 partial functions, including tri- parts A, B, C as shown in Figure 2, and it is respectively accuracy clothes that Each performs its own functions
Business.Wherein:
Part A is responsible for taking user's row data from database or caching, by analyzing different behaviors, generates current
The feature vector of user.If it is non-behavioural characteristic is used, does not just have to usage behavior and extract and analysis module, module output
Be user characteristics vector
Part B is responsible for converting initial recommendation columns of items by feature-article correlation matrix for the feature vector of user
Table.
C portion, be responsible for initial recommendation list is filtered, sequence processing, to generate consequently recommended result.
Each different part is explained in detail separately below:
1, user characteristics vector.Two kinds are generally comprised, one is that can be extracted from user's registration information, is mainly wrapped
Include the Demographics of user.It include: age, gender, nationality and the nationality etc. of user.For using pushing away for this feature
It recommends and holds up, if memory is enough, these characteristic informations can be stored directly in the caching of memory, system to be recommended takes user
Feature vector is generated after characteristic.In addition to this feature, another feature is mainly calculated from the behavior of user,
When calculating user behavior characteristics vector, design usually requires to consider to classify, more manageability after increasing convenient for late feature vector, this
Secondary design is classified by the type of user behavior, the time of user behavior generation, user behavior number:
The type of user behavior.In the end PC or APP, user can generate many different types of behaviors to product.With
Family may browse through product, click product link, collection product, give a mark to product, purchase, comment on, is tagged to product, becoming better
The different keyword etc. of friend's sharing product, search.These behaviors can all have an impact the weight of product feature, different behaviors
Influence different, Many times are difficult to determine that behavior is more important, and general standard is exactly that user pays a price bigger row
It is higher for weight.For example, purchase product user needs are footed a bill, so user is bound to think thrice before acting, therefore buying behavior is most
It is important.On the contrary, browsing product web page cost very little, so this behavior influences very little to the true interest of reflection user.
The time that user behavior generates.Ordinary circumstance, the recent behavior of user is important, and user for a long time before row
It is relatively secondary.Therefore, if certain product that user bought recently, the corresponding feature of this product will have ratio
Higher weight
The number of user behavior.Sometimes user can generate behavior many times to a product.For example user is subscribing freely
Row product can check repeatedly the strategy of free walker product repeatedly relatively with the free walker product of destination.Therefore user couple
The number occurred with the same behavior of product also reflects user to the interest of product, the corresponding spy of the product of behavior often
It is higher to levy weight.
2, feature-product associated recommendation
After obtaining user characteristics vector, according to initial recommendation result is calculated in part B function with correlation table.From
Line correlation table is typically stored in MySQL, and storage format is as follows:
src_id | product_id | distance_score |
Characteristic ID | Product IDs | Score value |
When further obtaining multiple features-product correlation table, calculate like product according to user characteristics, two vectors it
Between calculate distance, the nearlyr similarity of distance is bigger.User characteristics calculation formula is as follows:
Above-mentioned formula feature-product is mapped to a two-dimensional matrix, and Xi can be interpreted as to characteristic set src_id, Yi
For product product_id, vector calculates use as a unit to a product by all users for each possessing features described above
Similarity between family, and obtain distance_score score value and deposit table.It is all to obtain after experience is calculated with N number of like product
The product for obtaining score score value can all be enumerated;The product that he does not have preference and does not buy is predicted for active user simultaneously, is calculated
One sequence, provides and is given to initial recommendation result.Use is finally showed further according to filtering, the sequences such as range, limitation category can be sold
Family.
Fig. 3 is the schematic diagram data of above-mentioned formula.Left figure in Fig. 3 is that all users are calculating similarity with N number of product
Model_score afterwards, it be calculated by the calculating process and right figure of the middle figure two-dimensional matrix of Fig. 3 Lai score result
Demonstration.
Table is to take user A to traverse n product to obtain distance_ according to user characteristics calculation formula on the middle figure of Fig. 3
score.Review_score represents the feature of user.Here user comment feature has only been taken, in the actual production process also
Have the features such as the type of user behavior, the time that user behavior generates.
The middle figure following table of Fig. 3 be model_wt and distance_wt, review_wt and distance_ by each
Score and review_score does weighted calculation, and score points of right figure are generated after calculated weighted value normalizing fusion treatment
Value.
Many correlation tables can be configured in configuration file for a recommender system engine and be related to the weight of dimension, and
Online service can calculate these correlation tables according to the Weight of configuration when starting, and then be stored in final correlation table interior
It has been the correlation table after weighting in depositing, and when recommending to user.
3, candidate products set
The purpose of candidate products set is that guarantee recommendation results includes the product in candidate products set.Its applied field
Scape is usually artificial to wish certain Products Shows to can regularly publish the production cooperated with supplier on user, such as tour site
Product, it include this portioned product when giving user and recommending then the product of cooperation can be added in candidate products set.
4, filtering module
After obtaining preliminary recommendation list, this list can't be presented to user, need according to C portion function according to
Product demand is filtered result, filters out those undesirable products, in general, filtering module can filter out with
Lower product
User had generated buying behavior product intelligent recommender system and has been to aid in user's discovery product, therefore was not necessarily to
To the product that user recommends him to buy, filtering has generated the product of buying behavior, can promote the accuracy of recommendation.
It waits supplementing with 0 inventory and the product of undercarriage in the presence of can sell on the product tour site that foreground can not be sold, Ta Menke
It can have the opportunity to be recommended to user according to the product of collaborative filtering these types, therefore there are in a database table
Filtering module is needed to be filtered the product of undercarriage or the product of 0 inventory according to practical business demand, to guarantee that platform is sold in face of all circles
What is sold is the user experience that can be sold.
For the more secondary product of certain quality in order to improve user experience, intelligent recommendation system recommends high-quality production to user
Product obtain history score data for most review wt, that is, lower product of user comment score value, by number < 2 of scoring
All feedings filtering module of (general comment is divided into 5 points) filters.
5, ranking module
Being exposed directly to user by filtered recommendation results, also there is no problem, but if carrying out some rows to them
Name, then can preferably promote user satisfaction, therefore, in intelligent recommendation system increase many different ranking submodules, under
Characteristic in face of disparate modules is introduced respectively:
Novelty ranking.Although filtering module before, which has filtered certain customers once, the product of behavior, ensure that
A degree of novelty, but in actual use can have user and browse and understand certain products and not go
For the case where.Since such case user behavior feedback data can not obtain, can only go to speculate by approximate calculation, one
Kind mode drops power to product popular in recommendation results.
Formula (1) indicates, if user generates many behaviors to a hot product i, there is very big Pui, that is, phase
When similar with hot product in hot product, unexpected product is similar with unexpected product, if user likes a popular product,
Collaborative filtering is also difficult to recommend a unexpected product to him.The purpose for being thought of as product drop power is to recommend them for user not
Known product, then similar with i product and should be also that maximum probability is known, therefore can drop than the product user of i hot topic
The weight of low this product, to influence novel degree
Diversity.Diversity is also the another index of intelligent recommendation system, and increasing diversity can allow recommendation results to cover
User interest as much as possible.Promoted multifarious method be recommendation results are divided into according to certain product Hu contents attribute it is several
Class, recommendation results for then selecting top ranked product form final in each category, for example be free walker product can be with
Classify by island trip, Qin Ziyou, honeymoon trip etc. to recommendation results, the product form that TOPN is then chosen under every kind of classification finally pushes away
Recommend result.
Time diversity.Time diversity all sees same recommendation knot to guarantee that user not carry out recommender system daily
Fruit, so needing to adjust recommendation results in real time when obtaining user's new behavior to meet the needs of user is nearest.Concrete mode is,
If user has real-time behavior, User action log system, which can work, to be taken behavioral data and is converted into new feature, then
Be converted to the maximally related product of new feature by feature-product correlation module, thus in recommendation list with regard to immediate response user most
The influence of new behavior.
User feedback.The most important part of ranking module is user feedback module, and user feedback module mainly passes through analysis
Before user and recommendation results interactive log, predict that user can be interested in which type of recommendation results.For example, it travels the mesh of net
Mark is to improve user to the clicking rate of recommended products result, predicts whether user can click using the feature in click model and pushes away
Recommend result.It can predict that user can or can not click product with following feature in the clicking rate prediction of recommender system:
1) the relevant feature of user, for example, the age, gender, liveness, before whether there is or not click behavior
2) the relevant feature of product, such as popularity, average score, product attribute
3) position of the product in recommendation list.The click of user and the design of user interface have very high correlation, therefore
It is critically important whether position of the product in recommendation list clicks prediction user
4) whether user clicked before and there is recommended products same recommendation to explain other recommendation results recklessly
5) whether user clicked before and other recommendation results of recommended products from same recommended engine are when click mould
After type off-line calculation, then model is loaded into memory online, uses linear model to promote the efficiency of on-line prediction.
It is worth noting that although foregoing teachings are by reference to several essences that detailed description of the preferred embodimentsthe present invention has been described creates
Mind and principle, it should be appreciated that, the invention is not limited to the specific embodiments disclosed, the division also unawareness to various aspects
Taste these aspect in feature cannot combine, it is this divide merely to statement convenience.The present invention is directed to cover appended power
Included various modifications and equivalent arrangements in the spirit and scope that benefit requires.
Claims (7)
1. a kind of intelligent recommendation system, which includes modules A, module B and module C,
Modules A converts behavioural characteristic after extracting from user behavior data library to user behavior, in conjunction with user attribute data
The feature vector of library generation user;
Module B generates feature-article associated recommendation matrix from multiple correlation tables, and the feature vector of user passes through feature-article phase
It closes and matrix is recommended to be converted into initial recommendation item lists;
Module C, be responsible for initial recommendation list is filtered, ranking processing, to generate consequently recommended result.
2. intelligent recommendation system according to claim 1, which is characterized in that the recommender system includes candidate products collection
It closes, which is supplied to feature-article associated recommendation matrix reference of module B, while being also provided to the mistake of module C
Reference is provided during filter.
3. intelligent recommendation system according to claim 2, which is characterized in that in module C also by user behavior feedback and
Goods attribute is supplied to ranking processing and is referred to.
4. intelligent recommendation system according to claim 1, which is characterized in that the recommendation process of the system includes:
1) according to user behavior data and user attribute data, user characteristics vector is obtained;
2) according to user characteristics vector and feature-product correlation matrix, initial recommendation product list is obtained;
3) initial recommendation list is filtered, the processing such as ranking, generates consequently recommended result.
5. intelligent recommendation system according to claim 4, which is characterized in that when calculating feature vector using user behavior
Need to consider following factor:
1) user behavior is all kinds of, including browses, clicks product link, collection product, purchase, comment;
2) time that user behavior generates, including the recent behavior of user;
3) user behavior number;
4) popular degree of product,
Following product is filtered when filtering:
1) user had generated the product of purchase,
2) product that foreground can not be sold,
3) the more secondary product of quality, including the lower product of evaluation index,
Consideration when ranking:
1) novelty,
2) product diversity,
3) time diversity.
6. intelligent recommendation system according to claim 1, which is characterized in that the intelligent recommendation system passes through UI interconnection
User, while intelligent recommendation system also passes through UI system access log system, log system is stored by User action log is
Tieback enters intelligent recommendation system again after system,
UI system is responsible for interacting to user's displayed web page and with user, and UI system can be each on UI by user by log system
Various kinds behavior record is planted into the log of user behavior,
Log can store in memory cache, also can store in the database, when intelligent recommendation system is needed with these rows
It when for log, is taken from memory cache or database, so as to further analyze user behavior, generates portion and carry
It rubs the recommendation list of user intent, finally returns to UI system demonstration again in UI system.
7. intelligent recommendation system according to claim 6, which is characterized in that UI system passes through data transfer mode for product
User is shown at the moment with product attribute, including is stood in PC, the station H5, wechat, miscellaneous recommendation position in APP terminal, meanwhile,
Affix product title, thumbnail, product introduction product attribute,
Intelligent recommendation system can be according to these product attributes of the target adjustment of recommendation, comprising: promote user to place an order, increase preferential rush
It sells label, increase tag along sort, be the product introduction information coloring recommended originally,
Meanwhile user can generate oneself interbehavior when seeing the product that recommender system is recommended on website, including click,
Browse, place an order the product, these are referred to as the data of user behavior,
At this point, User action log storage system is started to work, user behavior data is collected and stored, intelligence is finally provided to and pushes away
It recommends and is counted, modeled, used, export new proposed algorithm, product is presented to PC or APP all circles again according to algorithmic rule
On the UI of face, to realize the effect of personalized recommendation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811417462.8A CN109584003A (en) | 2018-11-26 | 2018-11-26 | Intelligent recommendation system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811417462.8A CN109584003A (en) | 2018-11-26 | 2018-11-26 | Intelligent recommendation system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109584003A true CN109584003A (en) | 2019-04-05 |
Family
ID=65924135
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811417462.8A Pending CN109584003A (en) | 2018-11-26 | 2018-11-26 | Intelligent recommendation system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109584003A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111932338A (en) * | 2020-08-05 | 2020-11-13 | 深圳市分期乐网络科技有限公司 | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and storage medium |
CN112116427A (en) * | 2020-09-22 | 2020-12-22 | 深圳市分期乐网络科技有限公司 | Commodity recommendation method and device, electronic equipment and storage medium |
CN112767076A (en) * | 2021-01-08 | 2021-05-07 | 上海义缘网络科技有限公司 | Customer channel drainage system based on big data recommendation algorithm |
CN113468421A (en) * | 2021-06-29 | 2021-10-01 | 平安信托有限责任公司 | Product recommendation method, device, equipment and medium based on vector matching technology |
CN117333203A (en) * | 2023-12-01 | 2024-01-02 | 广东付惠吧数据服务有限公司 | Member marketing platform combined with business marketing solution |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104794207A (en) * | 2015-04-23 | 2015-07-22 | 山东大学 | Recommendation system based on cooperation and working method of recommendation system |
CN105512183A (en) * | 2015-11-24 | 2016-04-20 | 中国科学院重庆绿色智能技术研究院 | Personalized recommendation method and system based on users' independent choice |
US20160342679A1 (en) * | 2014-01-16 | 2016-11-24 | Shanghai Fusion Management Software Co., Ltd. | Recommendation method and recommendation system applied to social network |
-
2018
- 2018-11-26 CN CN201811417462.8A patent/CN109584003A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160342679A1 (en) * | 2014-01-16 | 2016-11-24 | Shanghai Fusion Management Software Co., Ltd. | Recommendation method and recommendation system applied to social network |
CN104794207A (en) * | 2015-04-23 | 2015-07-22 | 山东大学 | Recommendation system based on cooperation and working method of recommendation system |
CN105512183A (en) * | 2015-11-24 | 2016-04-20 | 中国科学院重庆绿色智能技术研究院 | Personalized recommendation method and system based on users' independent choice |
Non-Patent Citations (2)
Title |
---|
张宏 等: "基于用户评分和共同评分项的协同过滤算法研究", 《计算机应用研究》 * |
张宜浩 等: "《基于半监督学习的个性化推荐算法研究》", 31 May 2016, 重庆:重庆大学出版社 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111932338A (en) * | 2020-08-05 | 2020-11-13 | 深圳市分期乐网络科技有限公司 | Commodity recommendation method, commodity recommendation device, commodity recommendation equipment and storage medium |
CN112116427A (en) * | 2020-09-22 | 2020-12-22 | 深圳市分期乐网络科技有限公司 | Commodity recommendation method and device, electronic equipment and storage medium |
CN112767076A (en) * | 2021-01-08 | 2021-05-07 | 上海义缘网络科技有限公司 | Customer channel drainage system based on big data recommendation algorithm |
CN113468421A (en) * | 2021-06-29 | 2021-10-01 | 平安信托有限责任公司 | Product recommendation method, device, equipment and medium based on vector matching technology |
CN117333203A (en) * | 2023-12-01 | 2024-01-02 | 广东付惠吧数据服务有限公司 | Member marketing platform combined with business marketing solution |
CN117333203B (en) * | 2023-12-01 | 2024-04-16 | 广东付惠吧数据服务有限公司 | Member marketing platform combined with business marketing solution |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | On both cold-start and long-tail recommendation with social data | |
CN109584003A (en) | Intelligent recommendation system | |
Burke et al. | Matching recommendation technologies and domains | |
Chakraborty et al. | Fashion recommendation systems, models and methods: A review | |
CN106600302A (en) | Hadoop-based commodity recommendation system | |
KR20140026932A (en) | System and method providing a suited shopping information by analyzing the propensity of an user | |
TW200844776A (en) | System and method for associative matching | |
CN104268292A (en) | Label word library update method of portrait system | |
Biswas et al. | A hybrid recommender system for recommending smartphones to prospective customers | |
Madadipouya et al. | A literature review on recommender systems algorithms, techniques and evaluations | |
CN109918563A (en) | A method of the book recommendation based on public data | |
Zhu et al. | Neural attentive travel package recommendation via exploiting long-term and short-term behaviors | |
Almahmood et al. | Issues and solutions in deep learning-enabled recommendation systems within the e-commerce field | |
CN108122153A (en) | Personalized recommendation method based on cloud computing tupe under e-commerce environment | |
Ping et al. | User consumption intention prediction in meituan | |
Agarwal et al. | Recommender system for surplus stock clearance | |
Li et al. | Review of Recommender Systems and Their Applications | |
Zheng et al. | Incorporating price into recommendation with graph convolutional networks | |
Fang et al. | Implicit feedback recommendation method based on user-generated content | |
Patil et al. | Providing highly accurate service recommendation for semantic clustering over big data | |
CN115062215A (en) | Multimedia content recommendation method, device and storage medium | |
Gahier et al. | Cross Domain Recommendation Systems using Deep Learning: A Systematic Literature Review | |
Stalidis et al. | Recommendation Systems for e-Shopping: Review of Techniques for Retail and Sustainable Marketing | |
CN113792220A (en) | Target object recommendation method and device, computer equipment and storage medium | |
Ding et al. | [Retracted] Clustering Merchants and Accurate Marketing of Products Using the Segmentation Tree Vector Space Model |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190405 |
|
RJ01 | Rejection of invention patent application after publication |