CN105511901B - A kind of App cold start-up recommended method based on mobile App operation list - Google Patents
A kind of App cold start-up recommended method based on mobile App operation list Download PDFInfo
- Publication number
- CN105511901B CN105511901B CN201510628037.3A CN201510628037A CN105511901B CN 105511901 B CN105511901 B CN 105511901B CN 201510628037 A CN201510628037 A CN 201510628037A CN 105511901 B CN105511901 B CN 105511901B
- Authority
- CN
- China
- Prior art keywords
- app
- keyword
- calculated
- information
- operation list
- 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.)
- Active
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of App based on mobile App operation list to be cold-started recommended method, relates generally to the content of information retrieval and recommender system field, includes the following steps: to obtain a large amount of App information and carry out keyword arrangement;Calculate the weighted undirected graph for indicating keyword relationship;It is calculated and is pushed in the weighted undirected graph according to App information to be pushed.App based on mobile App operation list of the invention is cold-started recommended method, after training data trains initial model, model can be under the premise of well recommend using operation list content, it is good to solve the problems, such as cold start-up, and it is with good expansibility, and overcomes the deficiency based on commending contents to a certain extent.
Description
Technical field
The present invention relates to information retrieval, figure calculates and recommender system field, in particular to it is a kind of by keyword calculating and
Information retrieval and figure calculate and the operation based on mobile App and content information completed solve App in recommender system recommend it is cold
The method of starting problem.
Background technique
The fast development of mobile Internet provides diversified on-line off-line service for masses, greatly enrich and
Facilitate daily life.Meanwhile the wilderness demand in social life, the mobile application with new function continue to bring out, and are
People's lives provide more convenient and fast service.When carrying out the popularization of App, in order to rapidly enter in the public visual field,
General App developer can formulate certain push strategy.But in current App push mode, for new App push
Cold start-up problem is not settled properly always.
Existing the methods of random recommendation method, mean value method, mode method and the information Entropy Method for being directed to cold start-up, is very
The individual demand that user is sacrificed in big degree is cost, and then alleviates cold start-up problem to a certain extent.Giving new user
When push, general strategy is will to be pushed to new user, the in this way premise in unknown subscriber's feature using most wide or most fiery App
Under have certain push effect.When pushing new App to user, App is pushed to most active or downloading App most
User can be than recommending preferably to improve recommendation effect at random in this way under the premise of the feature of unknown App.But with
On method can only be alleviation cold start-up problem to a certain extent, and cannot eradicate.
Meanwhile based on original rating matrix expand method, directly utilize user demographic information and project it is interior
Hold characteristic information to be added in original user-article matrix.In this way when having new user or new article, can also make
These row or column are not empty in matrix, so that can be implemented when matrix calculating, can continue to calculate similar users yet
Or similar article, finally complete recommendation.This algorithm is effective to the addition of new user and new article, but for superelevation dimension
Data, when the dimension of user or article are far longer than the dimension of the information of expansion, these expand information and are just not enough to describe newly
The feature of user or new article, thus at this time can not be good solve the problems, such as cold start-up.
There are also use building probability statistics model and the method combined with machine learning.The former is due to collecting probability item
Very big cost is spent when part information, so being rarely employed.The latter chooses no sufficient mathematics to the ratio of influence factor
Foundation, so being only intended to specific data set.
For App recommendation, although content-based recommendation good can solve the problems, such as cold start-up, by real
Border test, effect is differed with random recommendation and is no different, therefore can not directly be used.
Summary of the invention
To solve the above-mentioned problems, the present invention provides a kind of App based on mobile App operation list to be cold-started recommendation side
Method, under the premise of being able to use user and running list and carry out the recommendation of good quality, while very good solution to be for new
The problem of article is cold-started.
To achieve the above object, the technical solution of the present invention is as follows:
A kind of App cold start-up recommended method based on mobile App operation list, includes the following steps:
S10. a large amount of App information are obtained and carry out keyword arrangement;
S20. the weighted undirected graph for indicating the keyword relationship is calculated;
S30. it is calculated and is pushed in the weighted undirected graph according to App information to be pushed.
Further, step S10 the following steps are included:
S101. the information of a large amount of App is obtained, includes App description information in the information;
S102. the operation list for obtaining a large number of users trains library as the data used are calculated;
S103. it to each App acquired in step s101 and its description information, is calculated using keyword method for solving every
The keyword and weight of a App description information.The keyword method for solving includes TF-IDF(Term Frequency-
Inverse Document Frequency, term frequency-inverse document frequency), the methods of TextRank.
Further, step S20 the following steps are included:
S201. according to the operation list of user, all App frequent item sets wherein included are calculated;
S202. according to the keyword of calculated App frequent item set and each App, calculate all keywords it
Between correlation, obtain keyword frequent item set;
S203. using the word in the keyword frequent item set being calculated as node, the associated weights between word are as power
Reassemble into weighted undirected graph.
Further, step S30 the following steps are included:
S301. the keyword and weight for calculating the description information of App to be pushed calculate this these in weighted undirected graph
The maximally related keyword set of keyword;
S302. corresponding App collection is mapped out according to the keyword set being calculated, according to the correlation of keyword to App
It is ranked up, obtains the frequent episode App that maximally related App collection namely most probable occur;
S303. to the user's push App to be pushed for being mounted with the frequent episode App.
Beneficial effects of the present invention: after training data trains initial model, model can be in using operation list
Hold under the premise of well recommend, very good solution is cold-started problem, and is with good expansibility, and to a certain extent
Overcome the deficiency based on commending contents.Specific mainly includes the following contents:
1) when recommending new App, its potential frequent episode App can be calculated, so according to the description information content of this App
After recommended, can overcome the problems, such as the cold start-up in conventional recommendation systems;
2) during recommending new App, understand the word node in progressive updating weighted undirected graph, therefore have good
Scalability.
3) the operation list based on App is weighted the building of non-directed graph, and then the information extracted in operation list carries out
It is further to recommend, therefore can overcome the shortcomings of to be based only on commending contents to a certain extent.
Detailed description of the invention
Fig. 1 is the general frame figure that a kind of App for running list based on mobile App of the invention is cold-started recommended method;
Fig. 2 is the flow diagram that a kind of App for running list based on mobile App of the invention is cold-started recommended method.
Specific embodiment
For a further understanding of the present invention, the preferred embodiment of the invention is described below with reference to embodiment, still
It should be appreciated that these descriptions are only further explanation the features and advantages of the present invention, rather than to the claims in the present invention
Limitation.
The present invention provides a kind of App based on mobile App operation list to be cold-started recommended method, dependent on user's
App runs the description information of list and a large amount of App, carries out the conversion of App frequent episode to keyword frequent episode, and by keyword and
Its relationship weight is converted to weighted undirected graph, when to recommend new App, extracts the keyword set of App first, and use weighting nothing
The correlativity that keyword set is calculated to figure, finally completes the conversion of the frequent episode by keyword set to App, and then reach use
The purpose of frequent episode recommendation App.
Referring to attached drawing 1-2, one embodiment of the present of invention the following steps are included:
Step S10, obtains a large amount of App information from network and user runs list information, and calculates the pass of App description information
Keyword, while according to the frequent item set of operation list information calculating App.Specifically, further comprising the steps;
Step S101 obtains a large amount of App and its description information, because description information is to can be good at describing a App
Feature information;The operation list information of a large number of users is obtained simultaneously;
Step S102 segments the description information of App first, and uses TF-IDF (Term Frequency-
Inverse Document Frequency, term frequency-inverse document frequency), the methods of TextRank calculate its keyword and weight;
The operation list of step S103, user are similar to supermarket shopping model, use Apriori(association rule algorithm) class
The frequent item set and weight of wherein App are calculated like method;
The calculating of App frequent item set is the frequent item set and weight in order to further calculate keyword;
Step S20 is calculated between keyword according to the relationship of the keyword of each App in App between App frequent item set
Frequent item set and weight, and be expressed as the form of weighted undirected graph.Specifically, further comprising the steps;
The weight of the frequent episode keyword occurred inside step S201, first calculating App.It is by each keyword when calculating
Weight be directly multiplied and obtain;
Step S202 calculates the weight of the keyword in App frequent item set between App, and by the weight of frequent item set to pass
Keyword weight is weighted, that is, calculating the product of weight between each keyword first, it is frequent to be then multiplied by this App again
The weight of item;
Weight computed above is added, the correlativity between final keyword is obtained, as frequent by step S203
The weight of item keyword;
Calculated frequent episode keyword and its weight are combined into weighted undirected graph by step S204, and wherein word is as section
Point, side of the weight as figure between word;
The weighted undirected graph being calculated is used to carry out frequent n-th-trem relation n between keyword and calculates, and carry out keyword and
The mapping of App;
Step S30, for new App to be pushed away, description thereof information first calculates keyword set, then by this keyword set
Most similar keyword set is calculated into weighted undirected graph, and corresponding App is finally mapped out by similar keyword set.Specifically
, further comprise the steps;
Step S301 calculates its description information using LDA method first according to new App to be pushed away and its description information
Keyword set and weight;
Step S302 calculates relative keyword set into weighted undirected graph using this keyword set and weight;
Calculated related keyword word set and weight are mapped out its corresponding App, these App are in theory by step S303
It is the possibility frequent episode App of App to be pushed away in meaning;
When calculating the associative key of keyword set from weighted undirected graph, if certain keywords to be calculated do not exist
, can be using word as node in weighted undirected graph, it and other are in keyword set to be calculated and between the keyword in figure
Correlativity weight is side, is added in weighted undirected graph, realizes the scalability of weighted undirected graph, with to new addition word
Use gradually adjust their relationship weights between other words.
Step S40 is the prediction frequent item set of App to be pushed away for calculated App.Therefore to being mounted with these App's
User recommends App to be pushed away.
The above description of the embodiment is only used to help understand the method for the present invention and its core ideas.It should be pointed out that pair
For those skilled in the art, without departing from the principle of the present invention, the present invention can also be carried out
Some improvements and modifications, these improvements and modifications also fall within the scope of protection of the claims of the present invention.
Claims (1)
1. a kind of App based on mobile App operation list is cold-started recommended method, which comprises the steps of:
S10. a large amount of App information are obtained and carry out keyword arrangement, are specifically included:
S101. the information of a large amount of App is obtained, includes App description information in the information;
S102. the operation list for obtaining a large number of users trains library as the data used are calculated;
S103. to each App acquired in step s101 and its description information, each App is calculated using keyword method for solving
The keyword and weight of description information;
S20. the weighted undirected graph for indicating keyword relationship is calculated, is specifically included:
S201. according to the operation list of user, all App frequent item sets wherein included are calculated;
S202. according to the keyword of calculated App frequent item set and each App, the phase between all keywords is calculated
Guan Xing obtains keyword frequent item set;
S203. using the word in the keyword frequent item set being calculated as node, the associated weights between word are as weight group
At weighted undirected graph;
S30. it is calculated and is pushed in the weighted undirected graph according to App information to be pushed, specifically included:
S301. the keyword and weight for calculating the description information of App to be pushed calculate these keywords in weighted undirected graph
Maximally related keyword set;
S302. corresponding App collection is mapped out according to the keyword set being calculated, App is carried out according to the correlation of keyword
Sequence obtains the frequent episode App that maximally related App collection namely most probable occur;
S303. to the user's push App to be pushed for being mounted with the frequent episode App.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510628037.3A CN105511901B (en) | 2015-09-28 | 2015-09-28 | A kind of App cold start-up recommended method based on mobile App operation list |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510628037.3A CN105511901B (en) | 2015-09-28 | 2015-09-28 | A kind of App cold start-up recommended method based on mobile App operation list |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105511901A CN105511901A (en) | 2016-04-20 |
CN105511901B true CN105511901B (en) | 2019-04-16 |
Family
ID=55719908
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510628037.3A Active CN105511901B (en) | 2015-09-28 | 2015-09-28 | A kind of App cold start-up recommended method based on mobile App operation list |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105511901B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107870934B (en) * | 2016-09-27 | 2021-07-20 | 武汉安天信息技术有限责任公司 | App user clustering method and device |
CN108182201B (en) * | 2017-11-29 | 2020-06-30 | 有米科技股份有限公司 | Application expansion method and device based on key keywords |
CN108173936A (en) * | 2017-12-27 | 2018-06-15 | 百度在线网络技术(北京)有限公司 | News recommends method and apparatus |
CN108520017B (en) * | 2018-03-21 | 2019-09-10 | Oppo广东移动通信有限公司 | Application program recommended method, device, server and storage medium |
CN110348920A (en) * | 2018-04-02 | 2019-10-18 | 中移(杭州)信息技术有限公司 | A kind of method and device of recommended products |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102063433A (en) * | 2009-11-16 | 2011-05-18 | 华为技术有限公司 | Method and device for recommending related items |
CN103518007A (en) * | 2011-03-08 | 2014-01-15 | 氯工程公司 | Apparatus for electrolyzing sulfuric acid and method for electrolyzing sulfuric acid |
CN103533390A (en) * | 2012-11-15 | 2014-01-22 | Tcl集团股份有限公司 | Method and system for recommending television programs based on social network information |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2015146130A (en) * | 2014-02-03 | 2015-08-13 | 富士通株式会社 | Information processing apparatus and information processing apparatus management method |
-
2015
- 2015-09-28 CN CN201510628037.3A patent/CN105511901B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102063433A (en) * | 2009-11-16 | 2011-05-18 | 华为技术有限公司 | Method and device for recommending related items |
CN103518007A (en) * | 2011-03-08 | 2014-01-15 | 氯工程公司 | Apparatus for electrolyzing sulfuric acid and method for electrolyzing sulfuric acid |
CN103533390A (en) * | 2012-11-15 | 2014-01-22 | Tcl集团股份有限公司 | Method and system for recommending television programs based on social network information |
Also Published As
Publication number | Publication date |
---|---|
CN105511901A (en) | 2016-04-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105511901B (en) | A kind of App cold start-up recommended method based on mobile App operation list | |
CN103209342B (en) | A kind of introduce video popularity and the collaborative filtered recommendation method of user's interests change | |
CN102902691B (en) | Recommend method and system | |
Zhai et al. | Mapping the popularity of urban restaurants using social media data | |
CN103544632B (en) | A kind of cyber personalized recommendation method and system | |
WO2017071251A1 (en) | Information pushing method and device | |
CN109166017A (en) | Method for pushing, device, computer equipment and storage medium based on reunion class | |
CN105809558A (en) | Social network based recommendation method and apparatus | |
CN101271559A (en) | Cooperation recommending system based on user partial interest digging | |
CN103886090A (en) | Content recommendation method and device based on user favorites | |
CN103544623A (en) | Web service recommendation method based on user preference feature modeling | |
CN101482884A (en) | Cooperation recommending system based on user predilection grade distribution | |
CN102999588A (en) | Method and system for recommending multimedia applications | |
CN104835057A (en) | Method and device for obtaining consumption feature information of network user | |
CN105976229A (en) | Collaborative filtering algorithm based on user and project mixing | |
CN103106285A (en) | Recommendation algorithm based on information security professional social network platform | |
WO2014193700A1 (en) | Social media pricing engine | |
AU2014353157B2 (en) | Methods and apparatus to measure a cross device audience | |
CN103914743A (en) | On-line serial content popularity prediction method based on autoregressive model | |
CN109034960A (en) | A method of more inferred from attributes based on user node insertion | |
CN105338408B (en) | Video recommendation method based on time factor | |
CN105812844B (en) | A kind of the user advertising method for pushing and system of TV | |
CN105550275A (en) | Microblog forwarding quantity prediction method | |
Yao | Exploiting human mobility patterns for point-of-interest recommendation | |
JP2013105213A (en) | Information recommending device and method, and device and program |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |