CN105989074A - Method and device for recommending cold start through mobile equipment information - Google Patents
Method and device for recommending cold start through mobile equipment information Download PDFInfo
- Publication number
- CN105989074A CN105989074A CN201510070689.XA CN201510070689A CN105989074A CN 105989074 A CN105989074 A CN 105989074A CN 201510070689 A CN201510070689 A CN 201510070689A CN 105989074 A CN105989074 A CN 105989074A
- Authority
- CN
- China
- Prior art keywords
- user
- app
- mobile device
- recommendation
- interest tags
- 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.)
- Granted
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Transfer Between Computers (AREA)
Abstract
The invention discloses a method and a device for recommending cold start through mobile equipment information. The method comprises the following steps of: acquiring the mobile equipment information of a user, and acquiring a mobile equipment type of the user and information of all APPs (Applications) mounted on mobile equipment through an operation system of the mobile equipment of the user; generating a recommendation list for the user, wherein the step of generating the recommendation list for the user includes the sub-steps of: producing a first recommendation list based on collaborative filtering, generating the recommendation list by taking contents which are enjoyed by other users and are similar to the mobile equipment type of the user and/or the mounted APPs in the database as recommendation contents, or producing the recommendation list based on interest label mapping, explicitly mapping the Apps to one or more interest labels, then filtering the corresponding content according to each interest label as the recommendation content to generate a second recommendation list; and recommending the contents in the first recommendation list or the second recommendation list to the user. The invention further discloses the device for recommending cold start through the mobile equipment information.
Description
Technical field
The present invention relates to one, particularly a kind of method and apparatus being carried out recommending cold start-up by mobile device information.
Background technology
The cold start-up problem of commending system, refers to lack, for new custom system, interest the content recommendation effectively that enough data capture user.This problem is that commending system is at one of actual product application significant challenge.Numerous in the solution of this problem, a widely used method of class is had to be to encourage user to log in commending system by social networks (Social Network Service:SNS) account.Commending system can utilize the interest model of information (such as concern relation, friend relation, interest tags, the issue content etc.) initialising subscriber of user social contact network, thus effectively recommends.Application No. " 201410020292 ", the Chinese invention patent application of entitled " based on microblogging word cloud generation method and the access support system of Users' Interests Mining ", it is proposed that a kind of method as user interest keyword for the keyword being issued content by digging user.For another example it is published in " label propagation algorithm is in the application of microblog users interest graph " literary composition of in July, 2012 " programmer " magazine, the social networks describing a kind of user of utilization converges to the method with user the interest tags of user good friend or perpetual object, so contribute to solving the openness problem of user's own interests label, because all not interested label of a large number of users.
But in actual applications, a lot of users are because worrying privacy concern or disliking trouble and will not select to log in recommended products by social networks account.According to statistics, the news in some hot topics is recommended in class product, is less than 50% with what the main flow social networks accounts such as microblogging, QQ, wechat logged in all users.This allows for the cold start-up problem that existing method is difficult to thoroughly solve the new user of commending system.By contrast, to propose a kind of this method universality higher for the present invention, and cold starting effect be no less than digging user social network information on many users.
Content of the invention
The technical problem to be solved is the drawbacks described above for prior art, a kind of method and apparatus being carried out by mobile device information and recommending cold start-up is provided, utilize user's mobile device information to carry out digging user interest, thus thoroughly solve the cold start-up problem of the new user of commending system.
To achieve these goals, the invention provides a kind of method being carried out by mobile device information and recommending cold start-up, wherein, comprise the steps:
S100, the mobile device information obtaining user, obtained mobile device model and installation all APP information on the mobile device of this user by the operating system of the mobile device of this user；
S200, produce for the recommendation list of this user, comprising:
S201, based on collaborative filtering produce the first recommendation list, the content that other users similar with the APP of the mobile device model of this user and/or installation in database are liked as content recommendation generation the first recommendation list；Or
S202, map based on interest tags and produce the second recommendation list, App is explicitly mapped to one or more interest tags, then screens corresponding content as content recommendation generation the second recommendation list according to each interest tags；
S300, by the commending contents in described first recommendation list or the second recommendation list give this user.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, is the collaborative filtering method finding based on similar users based on collaborative filtering in described step S201, comprising:
S2011, screening is conventional distinction App；
S2012, choose common smart mobile phone type；
S2013, App and type are mapped to specific dimension；
S2014, for given user, from the mobile device information of this given user extract mobile device characteristic vector；
S2015, find vector distance and nearest K the user of this given user based on WeakAND algorithm；
S2016, add up the content that in this K user, clicking rate is the highest as content recommendation.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, in described step S2014, the mobile device characteristic vector of the given user of extraction includes:
S20141, each App is mapped to the dimension between [0, N-1]；
S20142, each type being mapped to the dimension between [N, M-1], the value of dimension corresponding with user's type is 1, and the value of other dimensions is 0；
S20143, the value of the corresponding dimension of App of user installation are the access times of the nearest specified number of days of this user, and the value of other dimensions is 0.
The above-mentioned method being carried out by mobile device information recommending cold start-up, wherein, is the collaborative filtering method based on APP and model information based on collaborative filtering in described step S201, comprising:
S201a, periodically to add up clicking rate in the user group of each App and conventional type respectively be that the high of Top K clicks on contents list；
S201b, given user's mobile device, the App being installed by this user's mobile device and type obtain corresponding described high click contents list respectively；
S201c, merge obtain described high according to the weight of corresponding App and type and click on contents list and using Top N therein as content recommendation.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, in described step S201c, the sequence of content recommendation i when merging described high click contents list uses equation below to calculate:
Score (i)=sum (wa)+wd
Wherein, a represents an App of user, and i occurs in the Top K high click contents list of a, and wa represents the weight that App a can represent user interest；D represents the type of user, and wd represents the weight that user's type can represent user interest.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, described step S202 maps generation recommendation list based on interest tags and includes:
S2021, to App mark interest tags；
S2022, given user's mobile device, obtain the described interest tags of corresponding A pp that this user's mobile device is installed；
S2023, using the content related to described interest tags as content recommendation.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, marks interest tags and includes in described step S2021:
S20211, the tag database setting up commending system self；
S20212, the label capturing each APP in APP application market by webpage capture technology；
S20213, the label mapping of the described APP capturing in described tag database.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, marks interest tags and includes in described step S2021:
S2021a, the overall interest tags Top M adding up all users；
S2021b, a given App, statistics is mounted with in the user group of this App, L interest tags of most popular Top；
S2021c, compare L interest tags of this Top and overall situation interest tags Top M, take out and be different from the interest tags as this App for the interest tags of this overall situation Top M.
The above-mentioned method carrying out recommendation cold start-up by mobile device information, wherein, adds up the overall interest tags Top M of all users in described step S2021a, the interest tags using equation below calculating user is vectorial:
Wherein, Ti represents the interest tags vector of i-th user action, and wi represents the weight of i-th user action.
In order to above-mentioned purpose is better achieved, present invention also offers a kind of method being carried out by mobile device information and recommending cold start-up, wherein, comprise the steps:
S100, the mobile device information obtaining user, obtained mobile device model and installation all APP information on the mobile device of this user by the operating system of the mobile device of this user；
S200, producing the first recommendation list to this user based on collaborative filtering, the content liking other users similar with the APP of the mobile device model of this user and/or installation in database is as content recommendation generation the first recommendation list；
S300, map based on interest tags and produce the second recommendation list to this user, App is explicitly mapped to one or more interest tags, then screens corresponding content as content recommendation generation the second recommendation list according to each interest tags；
S400, merge described first recommendation list and described second recommendation list, and rearrange recommendation order according to the weighted sum of wherein listed content recommendation, generate a preferred recommendation list；
S500, by the commending contents in described preferred recommendation list give this user.
In order to above-mentioned purpose is better achieved, present invention also offers a kind of device for carrying out recommending the method for cold start-up above by mobile device information.
The beneficial functional of the present invention is:
Recommend cold start-up to compare with utilizing the user social contact network information to carry out in prior art, the present invention directly utilizes the information of user's mobile device, it is not necessary to user's Unsolicited Grant, coverage rate 100%.And in actual applications, the effect of the present invention will be better than the recommendation cold start-up based on the user social contact network information, and either use time and the content clicking rate of user has all met or exceeded equal level.
Describe the present invention below in conjunction with the drawings and specific embodiments, but not as a limitation of the invention.
Brief description
Fig. 1 is the method flow diagram of one embodiment of the invention；
Fig. 2 is the collaborative filtering method flow chart based on similar users discovery of one embodiment of the invention；
Fig. 3 is the extraction mobile device characteristic vector schematic diagram of one embodiment of the invention；
Fig. 4 is the collaborative filtering method flow chart based on APP and model information of one embodiment of the invention；
Fig. 5 is that mapping based on interest tags of one embodiment of the invention produces recommendation list flow chart；
Fig. 6 is the method flow diagram of another embodiment of the present invention.
Wherein, reference
S100-S500, S2011-S2016, S201a-S201c, S2021-S2023 step
Detailed description of the invention
Below in conjunction with the accompanying drawings the structural principle and operation principle of the present invention is described in detail:
Seeing Fig. 1, Fig. 1 is the method flow diagram of one embodiment of the invention.The method carrying out recommendation cold start-up by mobile device information of the present invention, may be used for recommending any object can classified by interest tags such as article, picture, video, commodity, comprises the steps:
Step S100, the mobile device information obtaining user, obtained mobile device model and installation all APP information on the mobile device of this user by the operating system of the mobile device of this user, for example can obtain unit type by the API that mobile device operation system provides, the mobile device information such as the App installing, the App being currently running.The Mobile operating system of main flow has disclosed API (Application Program Interface: application programming interfaces) to obtain some basic facility informations for third party App in the market.After obtaining user's mandate, the access rights that third party App obtains these API obtain the information needing.Such as, at Android platform, third party App can obtain all APP information (installation kit name) of unit type and installation.On iOS platform, situation is somewhat more complex, and third party App can not directly obtain the App list installed in user's mobile device, but can be by inquiring about this equipment and whether install the mode of certain App and indirectly obtain this information.This method requires the id list of the built-in conventional iOS App of third party App, and these data are can to capture from the webpage version of App Store.
Step S200, generation, for the recommendation list of this user, further include steps of
Step S201, based on collaborative filtering produce recommendation list, the content liking other users similar with the APP of the mobile device model of this user and/or installation in database generates recommendation list as content recommendation, i.e. pass through unit type, the same unit types of information recommendation such as App are installed, the content that same App user clicks on, shares, collects is installed.This strategy does not require that commending system understands user's type and installs the user interest that App reflects, for a given user U, it is only necessary to recommends identical with U type or installs the content that the similar user of App likes.Concrete, this strategy has two kinds of conventional implementation methods.
Or, it is possible to using step S202, mapping generation recommendation list based on interest tags, different with Synergistic method, this method is only with App information, it is impossible to utilize model information.App is explicitly mapped to one or more interest tags, then screen corresponding content according to each interest tags and generate recommendation list as content recommendation, i.e. by setting up the mapping relations of App and interest tags, the user installing specific App is utilized to the interest tags content recommendation of mapping.For example, " journey is taken " for App, if this App is stamped the label (corresponding user is likely to a business people often going on business) such as " travel for commercial purpose ", " job market ", for the user being mounted with this App, we just can recommend marked the popular article of " travel for commercial purpose " and " job market " label.Automatically there is the application of comparative maturity to the method that article labels, therefore repeated no more.
Step S300, by the commending contents in described recommendation list give this user.
Seeing Fig. 2, Fig. 2 is the collaborative filtering method flow chart based on similar users discovery of one embodiment of the invention.Described step S201 for example can be the collaborative filtering method based on similar users discovery, a given user U, in order to obtain suitable content recommendation lists IU, need first to find type and App list and U to be installed than relatively similar user.Specifically may include:
Step S2011, screening is conventional distinction App, for example can be by collecting the APP mount message of the sufficiently large all users of application of certain customer volume, adding up the user installation ratio of each APP, this statistics can simulate the proportion of installation in all mobile interchange network users for all App substantially.First, filtering the App more than 50% for the proportion of installation, this class is substantially microblogging, wechat, Baidu's client this whole people App, to identification user personalized interest not too big help.Secondly, filter the App less than 1% for the proportion of installation, it is assumed that mobile interchange network users has 500,000,000, and 1% means the installation of 50,000, less than this numeral it is believed that excessively minority, there is no statistical significance；
Step S2012, choose common smart mobile phone type；
Step S2013, App and type are mapped to specific dimension；
Step S2014, for given user, from the mobile device information of this given user extract mobile device characteristic vector；
Step S2015, find vector distance and nearest K the user of this given user based on WeakAND algorithm；
Step S2016, add up the content that in this K user, clicking rate is the highest as content recommendation.
Wherein, the mobile device characteristic vector of the given user of extraction in described step S2014, the mobile device information table of each user can be shown as equipment characteristic vector VU of a N+M dimension, wherein N represents considered App species, and M represents the quantity of main flow smart mobile phone type.Specifically include:
Step S20141, the dimension each App being mapped between [0, N-1] (such as ctrip.com > 2340)；
Step S20142, each type is mapped to [a N, M-1] between dimension (such as millet 2 > 10012), so give the information of a mobile device, it is obtained with a corresponding vector, the value of dimension corresponding with user's type is 1, and the value of other dimensions is 0；
Step S20143, the value of the corresponding dimension of App of user installation are the access times of the nearest specified number of days of this user (for example nearest seven days), and the value of other dimensions is 0.
For example, for being mounted with " ctrip.com (2340) ", a millet 2 (10012) equipment of " understanding ball Supreme Being (3078) ", it is possible to obtain the vector described by Fig. 3.Wherein, the value of 0-2339 dimension is 0, the value of the 2340th dimension is 1, the value of 2341-3077 dimension is 0, and the value of the 3078th dimension is 3, and the value of 3079-10011 dimension is 0, the value of the 10012nd dimension is 1, the value of 10013-(N+M-1) dimension is 0, then the value of the 2340th dimension be the value of the 1st, the 3078th dimension be 3 to represent access times in seven days for the corresponding APP respectively, the value of the 10012nd dimension is that 1 to represent user's type be 10012 (corresponding millets 2).I.e. on this vector, the value of the corresponding dimension of user installation App is represented by the access times of user, and the value of the corresponding dimension of user's type is 1, and the value of other dimensions is all 0.
Equipment characteristic vector VU of a given user can be able to conveniently find the Top K vector most like with this vector.If do a calculating to all devices characteristic vector, amount of calculation is very big (the recommended products user of such as some main flows is through hundred million), not only can consume a large amount of resource that calculates, and postpone possibly to accept (needing several hour).The classical WeakAND algorithm of this problem can solve well, and in reality, minute level can be disposed.After finding similar user, the content the highest of clicking rate in the past several hour can be added up in these users as recommendation list IU.It should be noted that at present, the Mobile solution market of main flow often has the application of hundreds of thousands kind available for download.It is to say, the N in the method can reach the magnitude of hundreds of thousands in theory, the expense so calculating similar users can be bigger.In actual applications, in order to remove noise and reduce computing cost, those downloads king-sized whole people level application (such as wechat, microblogging, QQ, Baidu's map etc.) and the few especially application of download can be filtered out, actually participate in the App species of calculating for example about 10,000.
The equipment characteristic vector assuming user u is Vu, the characteristic vector that Top K vector is exactly and Vu is most like.And vector similitude typically can use the angle of vector distance or vector to weigh.WeakAND algorithm is conventionally used to quickly search most like TopK the term vector (similar document) of a term vector (inquiry document) in text retrieval field.Its main thought is, a given term vector V1, the vector distance of another term vector V2 and V1 has a upper limit, and this upper limit can be decomposed into the contribution upper limit sum of each word.The contribution upper limit of each word is can be precalculated, such as one word " national football team " weight of maximum inside all article term vectors is 0.2, so the contribution upper limit of this word is exactly 0.2*0.2=0.04, that is, it easily is calculated, two articles only comprising a common word " national football team ", vector distance is 0.04 to the maximum.Quickly calculate the method for the vector distance upper limit, given V1, the nearest vector of K vector distance of Top to be found according to this, just all candidate vector all need not be calculated a distance and then sort.Generally can safeguard the queue of an a length of K, for front K vector, can directly first be put in queue.From the beginning of the K+1 vector, check that the distance upper limit of each vector, whether less than the vector with V1 distance minimum in queue, if it is lower, just directly abandon, otherwise calculates actual distance again, sees whether less than the vector with V1 distance minimum in queue.If less than or be equal to, just abandon, if it does, just in queue and the minimum vector of V1 distance is replaced out.So for a lot of vectors, it is possible to save the calculating of distance, and have only to scan candidate vector one time.In the present embodiment, if App and mobile device model are treated as word, user's mobile device characteristic vector treats as term vector, it is possible to uses WeakAND algorithm and obtains other user's mobile device characteristic vectors of the vectorial most like K with user's mobile device, i.e. Top K vector rapidly.
Seeing Fig. 4, Fig. 4 is the collaborative filtering method flow chart based on APP and model information of one embodiment of the invention.Described step S201 is alternatively the collaborative filtering method based on APP and model information based on collaborative filtering, comprising:
It is that the high of Top K clicks on contents list that step S201a, regular (such as each hour) add up in the user group of the corresponding user of each App and each conventional type clicking rate within the past period respectively, then statistics is saved in database or based on (access speed is faster) in the caching server of internal memory；
Step S201b, given user's mobile device, the App being installed by this user's mobile device and type obtain corresponding described high click contents list respectively；
Step S201c, merge obtain described high according to the weight of corresponding App and type and click on contents list and using Top N therein as content recommendation.
Wherein, in described step S201c, it may be considered that the weight of different App, the sequence of content recommendation i when merging described high click contents list uses equation below to calculate:
Score (i)=sum (wa)+wd
Wherein, a represents an App of user, and i occurs in the Top K high click contents list of a, and wa represents the weight that this App a can represent user interest；D represents the type of user, and wd represents the weight that this user's type can represent user interest.With the score according to each content, Top K piece content can be chosen and return as recommendation results.
Seeing Fig. 5, Fig. 5 is that mapping based on interest tags of one embodiment of the invention produces recommendation list flow chart.Described step S202 maps generation recommendation list based on interest tags and may include steps of:
Step S2021, to App mark interest tags；
Step S2022, given user's mobile device, obtain the described interest tags of corresponding A pp that this user's mobile device is installed；
Step S2023, using the content related to described interest tags as content recommendation.
Wherein, described step S2021 marks interest tags and can adopt with the following method, comprising:
Step S20211, the tag database setting up commending system self；
Step S20212, the label capturing each APP in APP application market (such as pea pods) by webpage capture technology；
Step S20213, the label mapping of the described APP capturing in described tag database.Why need to map and be because that the label system of App application market may not consistent with commending system self.Such as one UEFA Champions League video App, in application market may only one of which " UEFA Champions League " label, and be likely not to have " UEFA Champions League " inside commending system but have the label of " international soccer ", " UEFA Champions League " just should be mapped as " international soccer " in this case.Mapping relations can be by manual maintenance, it is also possible to automatically safeguarded by system, since it is desired that the number of labels mapping is usually no more than 1,000, and not needing there is more understanding to each App, being more prone to so operating.
A kind of method of the App of calculating weight is to compare the difference of the Top K list of an App and full clicking rate TopK content of standing, if two list differences are bigger, illustrates that this App more can embody the personalized interest of user, and weight is bigger.It is similar to, it is also possible to the Top K list according to type and full station clicking rate Top K content compare the weight obtaining type.For example, it in the user of today's tops, is mounted with that the article that the user of " knowing daily paper " likes the article clicked on the highest with full station clicking rate has significant difference.Clicking rate article the highest in full station is often amusement Eight Diagrams and the article of society's intriguing story class, is mounted with the high-quality content of the user then preference minority of this App.Therefore, the interest expression weight of this App is compared very high with general instrumental App.Concrete, wa can be calculated by equation below
Wa=1-| Ia I |/K
Wherein Ia, I represent Top K clicking rate contents list and the full station clicking rate Top K contents list being mounted with App a user respectively, and | Ia I | represents the size that both of which is occured simultaneously.Similarly, it is also possible to the Top K list according to type and full station clicking rate Top K content compare weight wd obtaining type.
Therefore, described step S2021 marks interest tags and may be used without following method, comprising:
Step S2021a, the overall interest tags Top M adding up all users；
Step S2021b, a given App, statistics is mounted with in the user group of this App, L interest tags of most popular Top；
Step S2021c, compare L interest tags of this Top and overall situation interest tags Top M, take out and be different from the interest tags as this App for the interest tags of this overall situation Top M.
The method requires to excavate in advance the interest tags of each user, then gives an App, and statistics is mounted with in the user group of this App, K interest tags of most popular Top.The Top M interest tags (overall situation Top M) finally taking K interest tags of this Top and all users to count contrasts, and takes out and the overall situation different interest tags as this App of Top M.It in the case of each content recommendation existing interest tags, is also a more ripe technology by user behavior digging user interest tags.Typically use the behavioral data of recommendation service (for example by user in recording station, which content browsed, click on/collect/commented on which content), and the interest tags of user and the weight of each label is excavated according to the interest tags of content, the two together constitutes the interest tags vector of user.Concrete grammar is as follows:
A. setting weight w for every kind of user action act, such as clicking on and calculate 1 point, but browse and do not click on calculation-0.2 point, collection calculates 5 points；
B. give sequence of user actions [act1, act2 ..., act3], the interest tags vector of user is calculated as follows:
Wherein Ti represents the interest tags vector of i-th user action, and wi represents the weight of i-th user action.
Seeing Fig. 6, Fig. 6 is the method flow diagram of another embodiment of the present invention.In the present embodiment, the method carrying out recommending cold start-up by mobile device information, it may include following steps:
Step S100, the mobile device information obtaining user, obtained mobile device model and installation all APP information on the mobile device of this user by the operating system of the mobile device of this user；
Step S200, producing the first recommendation list to this user based on collaborative filtering, the content liking other users similar with the APP of the mobile device model of this user and/or installation in database is as content recommendation generation the first recommendation list；
Step S300, map based on interest tags and produce the second recommendation list to this user, App is explicitly mapped to one or more interest tags, then screens corresponding content as content recommendation generation the second recommendation list according to each interest tags；
Step S400, merge described first recommendation list and described second recommendation list, and rearrange recommendation order according to the weighted sum of wherein listed content recommendation, generate a preferred recommendation list；
Step S500, by the commending contents in described preferred recommendation list give this user.
The method fully utilizes based on collaborative filtering and the advantage mapping two kinds of methods based on interest tags, mixes the recommendation results of the two.Concrete, the score of a content recommendation can be by the weighted sum being expressed as score in two kinds of strategies, and the content that such both of which is tended to recommend will obtain higher weight preferential recommendation to user.Its essence is exactly the consolidation problem of two score lists, can have a lot of method on processing.For example, the score of a content is made:
Score (i)=w1*score_1 (i)+w2*score_2 (i)
Wherein score_1 represents the score in list 1 for the i, and score_2 represents that the score in list 2 for the i, weight w1 and w2 can by experimental debugging out.
The present invention also provides a kind of device for carrying out recommending the method for cold start-up above by mobile device information, including data acquisition module, data processing module and data outputting module, wherein, data acquisition module, for obtaining the mobile device information of user, is obtained mobile device model and installation all APP information on the mobile device of this user by the operating system of the mobile device of this user；Data processing module is for processing the mobile device information of described data collecting module collected and generating recommendation list, including produce the first recommendation list to this user based on collaborative filtering, the content liking other users similar with the APP of the mobile device model of this user and/or installation in database generates the first recommendation list as content recommendation；Or map, based on interest tags, the second recommendation list producing to this user, App is explicitly mapped to one or more interest tags, then screens corresponding content according to each interest tags and generate the second recommendation list as content recommendation；And merge described first recommendation list and described second recommendation list, and rearrange recommendation order according to the weighted sum of wherein listed content recommendation, generate a preferred recommendation list；Data outputting module is used for the preferred content recommendation in the recommendation list that propelling data processing module generates to relative users.
The present invention directly utilizes the information of user's mobile device, it is not necessary to user's Unsolicited Grant, coverage rate 100%.And in actual applications, the effect of the present invention is better than the recommendation cold start-up based on the user social contact network information, and either use time and the content clicking rate of user is all up higher at least equal level.
Certainly; the present invention also can have other various embodiments; in the case of without departing substantially from present invention spirit and essence thereof; those of ordinary skill in the art work as and can make various corresponding change and deformation according to the present invention, but these change accordingly and deform the protection domain that all should belong to appended claims of the invention.
Claims (10)
1. one kind by mobile device information carry out recommend cold start-up method, it is characterised in that include as
Lower step:
S100, the mobile device information obtaining user, obtained by the operating system of the mobile device of this user
The mobile device model of this user and installation all APP information on the mobile device；
S200, produce for the recommendation list of this user, comprising:
S201, based on collaborative filtering produce the first recommendation list, by the movement with this user in database
The content that other similar users of the APP of unit type and/or installation like generates first as content recommendation
Recommendation list；Or
S202, based on interest tags map produce the second recommendation list, App is explicitly mapped to one
Individual or multiple interest tags, then screen corresponding content according to each interest tags and generate as content recommendation
Second recommendation list；
S300, by the commending contents in described first recommendation list or described second recommendation list give this user.
2. the method being carried out by mobile device information recommending cold start-up as claimed in claim 1, it is special
Levy and be, described step S201 is the collaborative filtering method finding based on similar users based on collaborative filtering,
Including:
S2011, screening is conventional distinction App；
S2012, choose common smart mobile phone type；
S2013, App and type are mapped to specific dimension；
S2014, for given user, from the mobile device information of this given user, extract mobile device special
Levy vector；
S2015, find vector distance and nearest K the user of this given user based on WeakAND algorithm；
S2016, add up the content that in this K user, clicking rate is the highest as content recommendation.
3. the method being carried out by mobile device information recommending cold start-up as claimed in claim 2, it is special
Levying and being, in described step S2014, the mobile device characteristic vector of the given user of extraction includes:
S20141, each App is mapped to the dimension between [0, N-1]；
S20142, each type is mapped to the dimension between [N, M-1], dimension corresponding with user's type
Value be 1, the value of other dimensions is 0；
S20143, the value of the corresponding dimension of App of user installation are the use time of the nearest specified number of days of this user
Number, the value of other dimensions is 0.
4. the method being carried out by mobile device information recommending cold start-up as claimed in claim 1, it is special
Levy and be, based on collaborative filtering for based on the collaborative filtering side of APP and model information in described step S201
Method, comprising:
S201a, periodically to add up clicking rate in the user group of each App and conventional type respectively be Top K
High click on contents list；
S201b, given user's mobile device, the App being installed by this user's mobile device and type are respectively
Obtain corresponding described high click contents list；
S201c, merge obtain described high according to the weight of corresponding App and type and click on contents list simultaneously
Using Top N therein as content recommendation.
5. the method being carried out by mobile device information recommending cold start-up as claimed in claim 3, it is special
Levying and being, in described step S201c, the sequence of content recommendation i when merging described high click contents list is adopted
Calculate by equation below:
Score (i)=sum (wa)+wd
Wherein, a represents an App of user, and i occurs in the Top K high click contents list of a
In, wa represents the weight that App a can represent user interest；D represents the type of user, and wd represents use
Family type can represent the weight of user interest.
6. the method being carried out by mobile device information recommending cold start-up as claimed in claim 1, it is special
Levying and being, described step S202 maps generation the second recommendation list based on interest tags and includes:
S2021, to App mark interest tags；
S2022, given user's mobile device, obtain the described of corresponding A pp that this user's mobile device installs
Interest tags；
S2023, using the content related to described interest tags as content recommendation.
7. the method being carried out by mobile device information recommending cold start-up as claimed in claim 6, it is special
Levy and be, described step S2021 marks interest tags and includes:
S20211, the tag database setting up commending system self；
S20212, the label capturing each APP in APP application market by webpage capture technology；
S20213, the label mapping of the described APP capturing in described tag database.
8. the method being carried out by mobile device information recommending cold start-up as claimed in claim 6, it is special
Levy and be, described step S2021 marks interest tags and includes:
S2021a, the overall interest tags Top M adding up all users；
S2021b, a given App, statistics is mounted with in the user group of this App, most popular Top L
Individual interest tags；
S2021c, compare L interest tags of this Top and the overall situation interest tags Top M, take out be different from this
The interest tags of overall situation Top M is as the interest tags of this App.
9. one kind by mobile device information carry out recommend cold start-up method, it is characterised in that include as
Lower step:
S100, the mobile device information obtaining user, obtained by the operating system of the mobile device of this user
The mobile device model of this user and installation all APP information on the mobile device；
S200, produce the first recommendation list to this user based on collaborative filtering, by database with this user
Mobile device model and/or the content liked of similar other users of the APP of installation raw as content recommendation
Become the first recommendation list；
S300, map based on interest tags and produce the second recommendation list to this user, App is explicitly reflected
It is mapped to one or more interest tags, then screen corresponding content as in recommendation according to each interest tags
Hold and generate the second recommendation list；
S400, merge described first recommendation list and described second recommendation list, and according to wherein listed recommendation
The weighted sum of content rearranges recommendation order, generates a preferred recommendation list；
S500, by the commending contents in described preferred recommendation list give this user.
10. one kind is used for being carried out by mobile device information described in any one in the claims 1-9
Recommend the device of the method for cold start-up.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510067401 | 2015-02-09 | ||
| CN2015100674013 | 2015-02-09 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105989074A true CN105989074A (en) | 2016-10-05 |
| CN105989074B CN105989074B (en) | 2019-07-05 |
Family
ID=57041126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510070689.XA Active CN105989074B (en) | 2015-02-09 | 2015-02-11 | A kind of method and apparatus recommend by mobile device information cold start-up |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105989074B (en) |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106776859A (en) * | 2016-11-28 | 2017-05-31 | 南京华苏科技有限公司 | Mobile solution App commending systems based on user preference |
| CN106846094A (en) * | 2016-12-29 | 2017-06-13 | 广州优视网络科技有限公司 | A kind of method and apparatus for recommending application message based on application has been installed |
| CN107223341A (en) * | 2017-04-01 | 2017-09-29 | 深圳市智晟达科技有限公司 | The method and commending system for recommending video are used according to app |
| CN107967276A (en) * | 2016-10-19 | 2018-04-27 | 阿里巴巴集团控股有限公司 | A kind of method and apparatus of recommended |
| CN108009247A (en) * | 2017-11-30 | 2018-05-08 | 广州酷狗计算机科技有限公司 | Information-pushing method and device |
| 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 |
| CN108520017A (en) * | 2018-03-21 | 2018-09-11 | 广东欧珀移动通信有限公司 | Application program recommends method, apparatus, server and storage medium |
| CN108809987A (en) * | 2018-06-14 | 2018-11-13 | 广州任天游网络科技有限公司 | A kind of online game extension method based on big data analysis |
| CN109145280A (en) * | 2017-06-15 | 2019-01-04 | 北京京东尚科信息技术有限公司 | The method and apparatus of information push |
| CN109688458A (en) * | 2019-01-14 | 2019-04-26 | 四川长虹电器股份有限公司 | The implementation method of smart television cloud desktop operation system based on big data algorithm |
| CN110263053A (en) * | 2019-06-17 | 2019-09-20 | 浙江每日互动网络科技股份有限公司 | A kind of server obtaining mobile terminal portrait label based on mobile terminal data |
| CN110858231A (en) * | 2018-08-07 | 2020-03-03 | 北京京东尚科信息技术有限公司 | Article recommendation method and device |
| WO2020052039A1 (en) * | 2018-09-13 | 2020-03-19 | 清华大学 | Optimal social recommendation method and device under limited attention |
| CN110955820A (en) * | 2018-09-22 | 2020-04-03 | 北京微播视界科技有限公司 | Media information interest point recommendation method, device, server and storage medium |
| CN111310016A (en) * | 2018-12-11 | 2020-06-19 | 百度在线网络技术（北京）有限公司 | Label mining method, device, server and storage medium |
| CN111708952A (en) * | 2020-06-18 | 2020-09-25 | 小红书科技有限公司 | Label recommendation method and system |
| CN111814032A (en) * | 2019-04-11 | 2020-10-23 | 阿里巴巴集团控股有限公司 | Cold start recommendation method and device and electronic equipment |
| CN112596712A (en) * | 2020-12-28 | 2021-04-02 | 上海风秩科技有限公司 | Cold start interface design method, system, electronic equipment and storage medium |
| CN113407845A (en) * | 2021-07-14 | 2021-09-17 | 上海明略人工智能(集团)有限公司 | Method and device for information recommendation, electronic equipment and storage medium |
| CN113761361A (en) * | 2021-07-29 | 2021-12-07 | 深圳市思为软件技术有限公司 | House property information recommendation method and terminal equipment |
| CN114707075A (en) * | 2022-06-06 | 2022-07-05 | 荣耀终端有限公司 | Cold start recommendation method and device |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101937547A (en) * | 2010-09-15 | 2011-01-05 | 宇龙计算机通信科技(深圳)有限公司 | Software and/or software information pushing method, system, acquisition device, software shop service system and mobile terminal |
| CN101959179A (en) * | 2009-07-17 | 2011-01-26 | 华为技术有限公司 | Method for providing mobile terminal application program, and server and mobile terminal |
| CN102592223A (en) * | 2011-01-18 | 2012-07-18 | 卓望数码技术（深圳）有限公司 | Commodity recommending method and commodity recommending system |
| CN103354561A (en) * | 2013-06-28 | 2013-10-16 | 贵阳朗玛信息技术股份有限公司 | Pushing method, processing method, and pushing device for configuration information |
| CN103399967A (en) * | 2013-08-26 | 2013-11-20 | 百度在线网络技术（北京）有限公司 | Software recommending method and system and server |
| US20140143087A1 (en) * | 2011-06-17 | 2014-05-22 | Kt Corporation | In-app recommendation system and user terminal |
| CN104133878A (en) * | 2014-07-25 | 2014-11-05 | 百度在线网络技术（北京）有限公司 | User label generation method and device |
| CN104216998A (en) * | 2014-09-10 | 2014-12-17 | 广州金山网络科技有限公司 | Method and device for recommending application program and terminal equipment |
| CN104239571A (en) * | 2014-09-30 | 2014-12-24 | 北京奇虎科技有限公司 | Method and device for application recommendation |
-
2015
- 2015-02-11 CN CN201510070689.XA patent/CN105989074B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101959179A (en) * | 2009-07-17 | 2011-01-26 | 华为技术有限公司 | Method for providing mobile terminal application program, and server and mobile terminal |
| CN101937547A (en) * | 2010-09-15 | 2011-01-05 | 宇龙计算机通信科技(深圳)有限公司 | Software and/or software information pushing method, system, acquisition device, software shop service system and mobile terminal |
| CN102592223A (en) * | 2011-01-18 | 2012-07-18 | 卓望数码技术（深圳）有限公司 | Commodity recommending method and commodity recommending system |
| US20140143087A1 (en) * | 2011-06-17 | 2014-05-22 | Kt Corporation | In-app recommendation system and user terminal |
| CN103354561A (en) * | 2013-06-28 | 2013-10-16 | 贵阳朗玛信息技术股份有限公司 | Pushing method, processing method, and pushing device for configuration information |
| CN103399967A (en) * | 2013-08-26 | 2013-11-20 | 百度在线网络技术（北京）有限公司 | Software recommending method and system and server |
| CN104133878A (en) * | 2014-07-25 | 2014-11-05 | 百度在线网络技术（北京）有限公司 | User label generation method and device |
| CN104216998A (en) * | 2014-09-10 | 2014-12-17 | 广州金山网络科技有限公司 | Method and device for recommending application program and terminal equipment |
| CN104239571A (en) * | 2014-09-30 | 2014-12-24 | 北京奇虎科技有限公司 | Method and device for application recommendation |
Non-Patent Citations (1)
| Title |
|---|
| 周伟华: "基于个性化推荐的移动阅读服务系统的研究与设计", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107967276A (en) * | 2016-10-19 | 2018-04-27 | 阿里巴巴集团控股有限公司 | A kind of method and apparatus of recommended |
| CN106776859A (en) * | 2016-11-28 | 2017-05-31 | 南京华苏科技有限公司 | Mobile solution App commending systems based on user preference |
| CN106846094A (en) * | 2016-12-29 | 2017-06-13 | 广州优视网络科技有限公司 | A kind of method and apparatus for recommending application message based on application has been installed |
| WO2018121700A1 (en) * | 2016-12-29 | 2018-07-05 | 广州优视网络科技有限公司 | Method and device for recommending application information based on installed application, terminal device, and storage medium |
| WO2018176454A1 (en) * | 2017-04-01 | 2018-10-04 | 深圳市智晟达科技有限公司 | Method for recommending videos according to app usage, and recommendation system |
| CN107223341A (en) * | 2017-04-01 | 2017-09-29 | 深圳市智晟达科技有限公司 | The method and commending system for recommending video are used according to app |
| CN109145280A (en) * | 2017-06-15 | 2019-01-04 | 北京京东尚科信息技术有限公司 | The method and apparatus of information push |
| CN108009247A (en) * | 2017-11-30 | 2018-05-08 | 广州酷狗计算机科技有限公司 | Information-pushing method and device |
| 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 |
| CN108520017A (en) * | 2018-03-21 | 2018-09-11 | 广东欧珀移动通信有限公司 | Application program recommends method, apparatus, server and storage medium |
| CN108809987A (en) * | 2018-06-14 | 2018-11-13 | 广州任天游网络科技有限公司 | A kind of online game extension method based on big data analysis |
| CN108809987B (en) * | 2018-06-14 | 2020-10-23 | 江苏果米文化发展有限公司 | Online game popularization method based on big data analysis |
| CN110858231A (en) * | 2018-08-07 | 2020-03-03 | 北京京东尚科信息技术有限公司 | Article recommendation method and device |
| WO2020052039A1 (en) * | 2018-09-13 | 2020-03-19 | 清华大学 | Optimal social recommendation method and device under limited attention |
| CN110955820A (en) * | 2018-09-22 | 2020-04-03 | 北京微播视界科技有限公司 | Media information interest point recommendation method, device, server and storage medium |
| CN111310016B (en) * | 2018-12-11 | 2023-08-04 | 百度在线网络技术（北京）有限公司 | Label mining method, device, server and storage medium |
| CN111310016A (en) * | 2018-12-11 | 2020-06-19 | 百度在线网络技术（北京）有限公司 | Label mining method, device, server and storage medium |
| CN109688458A (en) * | 2019-01-14 | 2019-04-26 | 四川长虹电器股份有限公司 | The implementation method of smart television cloud desktop operation system based on big data algorithm |
| CN111814032A (en) * | 2019-04-11 | 2020-10-23 | 阿里巴巴集团控股有限公司 | Cold start recommendation method and device and electronic equipment |
| CN111814032B (en) * | 2019-04-11 | 2024-05-28 | 阿里巴巴集团控股有限公司 | Cold start recommendation method and device and electronic equipment |
| CN110263053A (en) * | 2019-06-17 | 2019-09-20 | 浙江每日互动网络科技股份有限公司 | A kind of server obtaining mobile terminal portrait label based on mobile terminal data |
| CN111708952A (en) * | 2020-06-18 | 2020-09-25 | 小红书科技有限公司 | Label recommendation method and system |
| CN111708952B (en) * | 2020-06-18 | 2023-10-20 | 小红书科技有限公司 | Label recommending method and system |
| CN112596712A (en) * | 2020-12-28 | 2021-04-02 | 上海风秩科技有限公司 | Cold start interface design method, system, electronic equipment and storage medium |
| CN113407845A (en) * | 2021-07-14 | 2021-09-17 | 上海明略人工智能(集团)有限公司 | Method and device for information recommendation, electronic equipment and storage medium |
| CN113761361A (en) * | 2021-07-29 | 2021-12-07 | 深圳市思为软件技术有限公司 | House property information recommendation method and terminal equipment |
| CN114707075A (en) * | 2022-06-06 | 2022-07-05 | 荣耀终端有限公司 | Cold start recommendation method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105989074B (en) | 2019-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105989074A (en) | Method and device for recommending cold start through mobile equipment information | |
| CN104090919B (en) | Advertisement recommending method and advertisement recommending server | |
| CN107451861B (en) | Method for identifying user internet access characteristics under big data | |
| CN106168953B (en) | Bo-Weak-relationship social network-oriented blog recommendation method | |
| CN106789598B (en) | Social relation chain-based public number message pushing method, device and system | |
| CN104866969A (en) | Personal credit data processing method and device | |
| CN104394118A (en) | User identity identification method and system | |
| US20140201292A1 (en) | Digital business card system performing social networking commonality comparisions, professional profile curation and personal brand management | |
| CN110147821A (en) | Targeted user population determines method, apparatus, computer equipment and storage medium | |
| US20170364931A1 (en) | Distributed model optimizer for content consumption | |
| US20160117328A1 (en) | Influence score of a social media domain | |
| CN103106285A (en) | Recommendation algorithm based on information security professional social network platform | |
| US20180307733A1 (en) | User characteristic extraction method and apparatus, and storage medium | |
| Ha et al. | An analysis on information diffusion through BlogCast in a blogosphere | |
| CN104142975B (en) | Microblog information promotion method, device and system | |
| CN111967914A (en) | User portrait based recommendation method and device, computer equipment and storage medium | |
| CN110020149A (en) | Labeling processing method, device, terminal device and the medium of user information | |
| CN102063678A (en) | Method and device for distributing gifts to net friends on line | |
| CN103177129A (en) | Internet real-time information recommendation and prediction system | |
| CN106327211A (en) | SCRM system based on social media and development method thereof | |
| CN104090908A (en) | Method and device for counting mean detention time in page group and generalizing content in website | |
| CN109635192A (en) | Magnanimity information temperature seniority among brothers and sisters update method and platform towards micro services | |
| CN103870452A (en) | Method and method for recommending data | |
| Furini et al. | The use of hashtags in the promotion of art exhibitions | |
| CN109190027A (en) | Multi-source recommended method, terminal, server, computer equipment, readable medium |
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 | ||
| CP03 | Change of name, title or address | ||
| CP03 | Change of name, title or address |
Address after: Room 222, Floor 2, Building 1, Yard A23, North Third Ring West Road, Haidian District, Beijing 100098 Patentee after: Beijing Douyin Information Service Co.,Ltd. Address before: Room 10A, Block B, Yingdu Building, No. 48 Zhichun Road, Haidian District, Beijing 100041 Patentee before: BEIJING BYTEDANCE TECHNOLOGY Co.,Ltd. |