CN107563867A - A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit - Google Patents
A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit Download PDFInfo
- Publication number
- CN107563867A CN107563867A CN201710796663.2A CN201710796663A CN107563867A CN 107563867 A CN107563867 A CN 107563867A CN 201710796663 A CN201710796663 A CN 201710796663A CN 107563867 A CN107563867 A CN 107563867A
- Authority
- CN
- China
- Prior art keywords
- commodity
- user
- collection
- candidate
- data collection
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit, including:Carry out Data Collection structure commodity data collection and pre-process, obtain the commodity dominant character of format specification;According to commodity dominant character, based on potential Di Li Crays algorithm construction commodity stealth characteristics, the commodity stealth characteristics dimension of output is set, re-flags commodity;Candidate's commodity collection is built based on commodity data collection:Commodity data collection is clustered according to commodity stealth characteristics, by commercial articles clustering, the commodity in same class cluster have similar property, and the commercial variations in inhomogeneity cluster are larger, randomly select a commodity respectively from each class cluster, build candidate's commodity collection;Optimal commodity will be selected from candidate's commodity collection and are considered as multi-arm fruit machine problem, is calculated based on confidential interval upper bound algorithm and estimates a point highest commodity, as Recommendations;By the highest commercial product recommending that scored in candidate's commodity collection to user after, according to feedback renewal user characteristics and weight parameter.
Description
Technical field
The present invention relates to personalized recommendation technology, and in particular to a kind of commending system based on multi-arm fruit machine confidence upper limit
Cold start-up method.
Technical background
With the fast development of Information technology, internet all produces the data of magnanimity, people every day with the speed of explosion type
Produce, replicate, propagating the ability of information and greatly enhance, each user turns into the producer of internet information.User exists
Spend more and more time during the information for selecting oneself to need, or even can not independently screen at all, cause efficiency of information
Reduce, contain much information becomes a kind of burden on the contrary, problem of information overload occurs.Asked to preferably solve information overload
Topic, personalized recommendation system arise at the historic moment, and the system can accurately prejudge user's request according to user's history behavioural information, and then
Recommended.This is built upon a kind of Advanced Business intelligence system on the basis of mass data is excavated, and has provided the user completely
Personalized decision support and information service, makes people more benefit from internet, big data.
Personalized recommendation system is that structure is for particular user according to information such as the historical behavior of user and purchaser records
Personalized user feature, commodity are screened, recommend with user characteristics similar in commodity.At present, personalized recommendation system
The every field in internet is widely used.Such as the e-commerce fields such as Amazon, Taobao, today's tops etc.
The cinematographic field such as music field, Netflix, bean cotyledon such as News Field, Netease's cloud music is all using proposed algorithm.Individual character at present
Changing commending system method mainly has rule-based recommendation, collaborative filtering recommending, content-based recommendation, the recommendation based on social activity
With mixing commending system etc..
The overall flow of personalized recommendation system mainly includes:Collect and arrange the historical record and behavior feedback structure of user
Build data set;User characteristics is obtained with corresponding algorithm according to data set;Corresponding commodity are chosen according to user characteristics, and
By commercial product recommending to user;Feedback of the user to Recommendations is recorded, recommendation effect is evaluated and updates the data collection.
The content of the invention
The purpose of the present invention is exactly that the confidential interval upper limit algorithm in multi-arm fruit machine model is applied into commending system
In cold start-up, recommendation process is considered as multi-arm fruit machine model, and unknown user is characterized based on confidential interval upper limit algorithm
Recommended, according to the click behavior of user, constantly fit and level off to the real feature of user, it is more next so as to be carried out for user
More accurate recommendation, solves the problems, such as cold start-up.
A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit, comprises the following steps:
(1) carry out Data Collection structure commodity data collection and pre-process, obtain the commodity dominant character of format specification:From net
Network platform chooses a number of commodity, forms commodity data collection, including commodity ID, user and businessman are beaten the commodity
Label, evaluation information, commodity text message is pre-processed, the commodity dominant character of output format specification;
(2) according to commodity dominant character, based on potential Di Li Crays algorithm construction commodity stealth characteristics, the business of output is set
Product stealth characteristics dimension, re-flags commodity ID;
(3) initialising subscriber feature and relevant parameter:In cold start-up problem, user characteristics is unknown, it is necessary to special to user
Sign and corresponding weight parameter assign initial value;
(4) based on commodity data collection structure candidate's commodity collection:K- is carried out to commodity data collection according to commodity stealth characteristics
Means is clustered, and by commercial articles clustering, the commodity in same class cluster have similar property, the commercial variations in inhomogeneity cluster
Property it is larger, randomly select a commodity respectively from each class cluster, build candidate's commodity collection;
(5) optimal commodity will be selected from candidate's commodity collection and is considered as multi-arm fruit machine problem, calculated based on the confidential interval upper bound
Method, which calculates, estimates a point highest commodity, as Recommendations;
(6) by the highest commercial product recommending that scored in candidate's commodity collection to user after, user will like according to itself, and selection be
No click commodity, so as to which system obtains feedback of the user to recommendation results, joined according to feedback renewal user characteristics and weight
Number;
(7) above-mentioned (3) are repeated and arrives (6) step, more wheels are carried out to user and are recommended, according to more wheel feedback results, constantly updated
User characteristics, so as to reach higher and higher user characteristics degree of fitting, so as to obtain accurate recommendation, alleviate cold start-up and ask
Topic.
Embodiment
The present invention introduces the thought of multi-arm fruit machine model and confidential interval in the cold start-up problem of recommendation process.
The thought is to the effect that:Using commodity data collection known to feature, it is characterized unknown user and is recommended, according to user
Click behavior, constantly fit and level off to the real feature of user, so as to for user more and more accurately recommend, solve
Cold start-up problem.Comprise the following steps that:
1 data set pre-processes
A number of commodity are chosen from the network platform first, commodity selected by acquiescence are not new restocking commodity, therefore can be with
The information provided according to businessman, and evaluation of the user to commodity, classification obtain the dominant character of commodity.Then commodity are shown
Property feature is pre-processed, because user's word that labelled to commodity has non-standard phenomena in dominant character, such as by two words
Write the two or more syllables of a word together together, synonym, it is not intended to adopted mess code etc., therefore to carry out cutting word, take stem, deactivation etc. handle, it is most dominant at last
Feature is organized into the form of the keyword of specification, is easy to subsequent treatment.
2 calculate the potential feature of commodity, build commodity data collection
By the dominant character of the specification of each commodity, the stealth characteristics of commodity are calculated using potential Di Li Crays algorithm,
And intrinsic dimensionality is arranged to 25 dimensions, product features are represented with x, and x is the column vector of 25 dimensions.Extract the commodity weight after potential feature
New Tag ID, it is positioned in database with to be used.Wherein, the collection based on Python is provided in scikit-learn
, can be in the data input module that easily will be by pretreatment as the potential Di Li Crays algorithm of module, and export business
The potential feature of product.
3 initialising subscriber features and parameter
In the cold start-up problem of commending system, user characteristics is unknown, it is therefore desirable to first presets an initial value, uses
Family feature is represented with θ.Here also need to define two parameters, 25 dimension matrix As and 25 dimensional vector b, user characteristics θ can be by as follows
Formula represents:
θ=A-1b
A initial value is arranged to unit matrix in formula, and b initial value is arranged to null vector.
4 select candidate's commodity from commodity data concentration
K-means clusters are carried out to commodity data collection according to potential feature, commodity are divided into 100 class clusters, in same class
Commodity in cluster have similar property, and the commercial variations in inhomogeneity cluster are larger.Every time from 100 class clusters respectively with
Machine extracts a commodity, therefore can ensure that 100 commodity are mutually dissimilar.And randomly select out 100 candidate's commodity are put
Enter in candidate's commodity collection.
5 calculate optimal commodity from candidate's commodity collection
The candidate's commodity amount times deposited in candidate's commodity collection is so larger, and whole candidate's commodity are recommended into user and just lost
Improved effect, it is therefore desirable to further filter out user's most probable commodity interested and recommended, recommended time minimum
The potential feature of user is fitted under several.Therefore using confidential interval upper bound algorithm in multi-arm fruit machine model, known users are liked
The joyous type of merchandise carries out exploitation recommendation more, and tera incognita is attempted on a small quantity, constantly explores, is converted into the form of mathematics just
It is to be given a mark according to the commodity that are fed back to of the number according to trial and user, commodity fraction represents that formula is as follows with p:
In formula, x calculates the stealth characteristics of commodity using potential Di Li Crays algorithm in being 5.1.
100 candidate's commodity are calculated with user respectively and scored, and by commodity according to the arrangement scored from high to low, selection
Wherein scoring highest commodity are recommended user.
6 users are fed back and update user characteristics and parameter
By the highest commercial product recommending that scored in candidate's commodity collection to user after, user will like according to itself, that is, latent
In feature, choose whether to click on the commodity, introduce clicking rate r herein to record the operation behavior of user, clicking rate r is Boolean type
Parameter, then r=1 is clicked on, does not click on then r=0.After obtaining user feedback, then parameter A and b is updated, more new formula is such as
Under:
A=A+xxT
B=b+rx
7 repeat to recommend, and are fitted to user characteristics
The step in the above 3 to 6 is repeated, that is, is repeated as user and is recommended, user's type of merchandise quilt interested
The probability of recommendation will constantly increase, and user is not apparent from showing that the type of merchandise interested will be to recommend, for exploring compared with small probability
User interest, user repeatedly show that the recommended probability of the uninterested type of merchandise will be more and more lower.Repeat take second place more
Afterwards, the potential feature of user will be accurately fitted, and cold start-up problem is effectively solved.
Claims (1)
1. a kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit, comprises the following steps:
(1) carry out Data Collection structure commodity data collection and pre-process, obtain the commodity dominant character of format specification:Put down from network
Platform chooses a number of commodity, forms commodity data collection, including commodity ID, user and businessman are to the commodity institute mark
Label, evaluation information, are pre-processed, the commodity dominant character of output format specification to commodity text message;
(2) according to commodity dominant character, based on potential Di Li Crays algorithm construction commodity stealth characteristics, set the commodity of output hidden
Shape characteristic dimension, re-flag commodity ID;
(3) initialising subscriber feature and relevant parameter:In cold start-up problem, user characteristics it is unknown, it is necessary to user characteristics with
And corresponding weight parameter assigns initial value;
(4) based on commodity data collection structure candidate's commodity collection:K-means is carried out according to commodity stealth characteristics to commodity data collection to gather
Class, by commercial articles clustering, the commodity in same class cluster have similar property, and the commercial variations in inhomogeneity cluster are larger,
Randomly select a commodity respectively from each class cluster, build candidate's commodity collection;
(5) optimal commodity will be selected from candidate's commodity collection and is considered as multi-arm fruit machine problem, based on confidential interval upper bound algorithm meter
Calculate and estimate a point highest commodity, as Recommendations;
(6) by candidate's commodity collection score highest commercial product recommending to user after, user will according to itself like, choose whether a little
The commodity are hit, so as to which system obtains feedback of the user to recommendation results, according to feedback renewal user characteristics and weight parameter;
(7) above-mentioned (3) are repeated and arrives (6) step, more wheels are carried out to user and are recommended, according to more wheel feedback results, constantly update user
Feature, so as to reach higher and higher user characteristics degree of fitting, so as to obtain accurate recommendation, alleviate cold start-up problem.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710796663.2A CN107563867A (en) | 2017-09-06 | 2017-09-06 | A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710796663.2A CN107563867A (en) | 2017-09-06 | 2017-09-06 | A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107563867A true CN107563867A (en) | 2018-01-09 |
Family
ID=60978147
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710796663.2A Pending CN107563867A (en) | 2017-09-06 | 2017-09-06 | A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107563867A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108595595A (en) * | 2018-04-19 | 2018-09-28 | 北京理工大学 | A kind of user knowledge requirement acquisition method calculated based on interactive differential evolution |
CN109949099A (en) * | 2019-03-23 | 2019-06-28 | 西安电子科技大学 | Information core construction method based on cluster and multi-arm fruit machine |
CN110335123A (en) * | 2019-07-11 | 2019-10-15 | 创新奇智(合肥)科技有限公司 | Method of Commodity Recommendation, system, computer-readable medium and device based on social electric business platform |
CN110348947A (en) * | 2019-06-13 | 2019-10-18 | 阿里巴巴集团控股有限公司 | Object recommendation method and device |
CN110766456A (en) * | 2019-10-16 | 2020-02-07 | 无线生活(杭州)信息科技有限公司 | Commodity recommendation method and device |
CN111582975A (en) * | 2020-04-23 | 2020-08-25 | 许立达 | Artificial intelligence recommendation method and system based on combination of users, products and advertisements |
CN111861605A (en) * | 2019-04-28 | 2020-10-30 | 阿里巴巴集团控股有限公司 | Business object recommendation method |
US10936961B1 (en) | 2020-08-07 | 2021-03-02 | Fmr Llc | Automated predictive product recommendations using reinforcement learning |
CN112733004A (en) * | 2021-01-22 | 2021-04-30 | 上海交通大学 | Movie and television work recommendation method based on multi-arm tiger machine algorithm |
CN117788105A (en) * | 2023-12-25 | 2024-03-29 | 公安县谦合广告装饰有限公司 | Online live broadcast method of E-commerce based on Internet |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104991966A (en) * | 2014-07-29 | 2015-10-21 | Tcl集团股份有限公司 | Ordering method and system of media content |
CN106326351A (en) * | 2016-08-05 | 2017-01-11 | 天津大学 | Recommendation system cold start solving method based on user feedback |
US20170083965A1 (en) * | 2014-06-10 | 2017-03-23 | Huawei Technologies Co., Ltd. | Item Recommendation Method and Apparatus |
-
2017
- 2017-09-06 CN CN201710796663.2A patent/CN107563867A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170083965A1 (en) * | 2014-06-10 | 2017-03-23 | Huawei Technologies Co., Ltd. | Item Recommendation Method and Apparatus |
CN104991966A (en) * | 2014-07-29 | 2015-10-21 | Tcl集团股份有限公司 | Ordering method and system of media content |
CN106326351A (en) * | 2016-08-05 | 2017-01-11 | 天津大学 | Recommendation system cold start solving method based on user feedback |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108595595A (en) * | 2018-04-19 | 2018-09-28 | 北京理工大学 | A kind of user knowledge requirement acquisition method calculated based on interactive differential evolution |
CN108595595B (en) * | 2018-04-19 | 2020-06-16 | 北京理工大学 | User knowledge demand acquisition method based on interactive differential evolution calculation |
CN109949099A (en) * | 2019-03-23 | 2019-06-28 | 西安电子科技大学 | Information core construction method based on cluster and multi-arm fruit machine |
CN109949099B (en) * | 2019-03-23 | 2022-04-08 | 西安电子科技大学 | Information core construction method based on clustering and multi-arm gambling machine |
CN111861605A (en) * | 2019-04-28 | 2020-10-30 | 阿里巴巴集团控股有限公司 | Business object recommendation method |
WO2020221022A1 (en) * | 2019-04-28 | 2020-11-05 | 阿里巴巴集团控股有限公司 | Service object recommendation method |
CN110348947B (en) * | 2019-06-13 | 2022-02-25 | 创新先进技术有限公司 | Object recommendation method and device |
CN110348947A (en) * | 2019-06-13 | 2019-10-18 | 阿里巴巴集团控股有限公司 | Object recommendation method and device |
CN110335123B (en) * | 2019-07-11 | 2021-12-07 | 创新奇智(合肥)科技有限公司 | Commodity recommendation method, system, computer readable medium and device based on social e-commerce platform |
CN110335123A (en) * | 2019-07-11 | 2019-10-15 | 创新奇智(合肥)科技有限公司 | Method of Commodity Recommendation, system, computer-readable medium and device based on social electric business platform |
CN110766456A (en) * | 2019-10-16 | 2020-02-07 | 无线生活(杭州)信息科技有限公司 | Commodity recommendation method and device |
CN111582975A (en) * | 2020-04-23 | 2020-08-25 | 许立达 | Artificial intelligence recommendation method and system based on combination of users, products and advertisements |
CN111582975B (en) * | 2020-04-23 | 2023-06-02 | 许立达 | Artificial intelligence recommendation method and system based on combination of user, product and advertisement |
US10936961B1 (en) | 2020-08-07 | 2021-03-02 | Fmr Llc | Automated predictive product recommendations using reinforcement learning |
CN112733004A (en) * | 2021-01-22 | 2021-04-30 | 上海交通大学 | Movie and television work recommendation method based on multi-arm tiger machine algorithm |
CN112733004B (en) * | 2021-01-22 | 2022-09-30 | 上海交通大学 | Movie and television work recommendation method based on multi-arm tiger machine algorithm |
CN117788105A (en) * | 2023-12-25 | 2024-03-29 | 公安县谦合广告装饰有限公司 | Online live broadcast method of E-commerce based on Internet |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107563867A (en) | A kind of commending system cold start-up method based on multi-arm fruit machine confidence upper limit | |
US11687992B2 (en) | System and method for providing food taxonomy based food search and recommendation | |
US11847684B2 (en) | Systems and method for tracking consumer tasting preferences | |
CN104866474B (en) | Individuation data searching method and device | |
CN103246980B (en) | Information output method and server | |
US8694362B2 (en) | Taxonomy based targeted search advertising | |
JP5386663B1 (en) | Information processing apparatus, information processing method, information processing program, and recording medium | |
CN110162706A (en) | A kind of personalized recommendation method and system based on interaction data cluster | |
CN101996215B (en) | Information matching method and system applied to e-commerce website | |
CN106651546A (en) | Intelligent community oriented electronic commerce information recommendation method | |
CN103473354A (en) | Insurance recommendation system framework and insurance recommendation method based on e-commerce platform | |
Ban et al. | Semantic network analysis of hotel package through the big data | |
WO2008012834A2 (en) | A method and a system for searching information using information device | |
CN107220365A (en) | Accurate commending system and method based on collaborative filtering and correlation rule parallel processing | |
US20200294071A1 (en) | Determining user intents related to websites based on site search user behavior | |
US20160299973A1 (en) | Interactive Internet Interfaces | |
CN106326318B (en) | Searching method and device | |
CN103514253A (en) | Ranking based on social activity data | |
Nilesh et al. | Recommendation of Indian cuisine recipes based on ingredients | |
CN103136257A (en) | Information providing method and device thereof | |
US10474670B1 (en) | Category predictions with browse node probabilities | |
JP5492160B2 (en) | Association apparatus, association method, and association program | |
CN114926199A (en) | Internet marketing audience accurate analysis method and system | |
Liao et al. | Mining information users’ knowledge for one-to-one marketing on information appliance | |
Bhojne et al. | Collaborative approach based restaurant recommender system using Naive Bayes |
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: 20180109 |
|
RJ01 | Rejection of invention patent application after publication |