CN110457581A - A kind of information recommended method, device, electronic equipment and storage medium - Google Patents
A kind of information recommended method, device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN110457581A CN110457581A CN201910712063.2A CN201910712063A CN110457581A CN 110457581 A CN110457581 A CN 110457581A CN 201910712063 A CN201910712063 A CN 201910712063A CN 110457581 A CN110457581 A CN 110457581A
- Authority
- CN
- China
- Prior art keywords
- information
- label
- recalled
- user
- vector
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention provides a kind of information recommended method, device, electronic equipment and storage mediums, wherein, this method comprises: determining the M information that the user clicks recently, and using the arithmetic average of the label vector of each information as corresponding information vector；Similarity based on the information vector of each information in the M information recalls K information；The linked character of each information recalled and each information recalled is input in the mathematical model trained, the clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate；Information recommendation is carried out to user based on ranking results.Technical solution provided in an embodiment of the present invention can make the information recalled have diversity and novelty, to increase the diversity and novelty for recommending information, can be user development more interested information, improve the clicking rate of information.
Description
Technical field
The present embodiments relate to data processing technique more particularly to a kind of information recommended method, device, electronic equipment and
Storage medium.
Background technique
With the fast development of Internet technology, more and more users by computer equipment (for example, smart phone,
PC etc.) obtain information information, but explosive growth is presented in information of today, massive information makes user be difficult to find
Information needed for oneself.Information miscarriage product come into being, and can precisely be distributed information according to the hobby of user
Push.
Information miscarriage product need user during constantly interacting with computer equipment, need to increase recommendation information
Novelty and diversity, better meet user interest explore demand, i.e., while meeting customer consumption relevent information,
The interested information of user can also be expanded.
Summary of the invention
The embodiment of the invention provides a kind of information recommended method, device, electronic equipment and storage medium, can make to recall
Information have diversity and novelty, thus increase recommend information diversity and novelty, can for user development more
Mostly interested information, improves the clicking rate of information.
In a first aspect, the embodiment of the invention provides a kind of information recommended methods, comprising:
Determine the M information that user clicks recently, and using the arithmetic average of the label vector of each information as correspondence
Information vector；
Similarity based on the information vector of each information in the M information recalls K information；
The linked character of each information recalled and each information recalled is input in the mathematical model trained,
The clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate；
Information recommendation is carried out to user based on ranking results.
Second aspect, the embodiment of the invention also provides a kind of information recommendation apparatus, comprising:
Information vector determining module, the M information clicked recently for determining user, and by the label vector of each information
Arithmetic average as corresponding information vector；
Module is recalled, recalls K information for the similarity based on the information vector of each information in the M information；
Clicking rate determining module, for the linked character of each information recalled and each information recalled to be input to
In the mathematical model trained, the clicking rate for each information recalled；
Recommending module for being ranked up based on the clicking rate to all information recalled, and is based on ranking results pair
User carries out information recommendation.
The third aspect, the embodiment of the invention also provides a kind of electronic equipment, comprising:
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device realizes a kind of information recommended method provided in an embodiment of the present invention.
A kind of fourth aspect, computer readable storage medium provided in an embodiment of the present invention, is stored thereon with computer journey
Sequence, the program realize a kind of information recommended method provided in an embodiment of the present invention when being executed by processor.
Technical solution provided in an embodiment of the present invention passes through the mark of each information in M information clicking user recently
The arithmetic mean of instantaneous value of vector is signed as corresponding information vector, and the similarity based on information vector recalls K information, and will call together
It the information returned and recalls the linked character of each information and is input in mathematical model, the click for each information recalled
Rate, and be ranked up information based on clicking rate, information recommendation is carried out to user based on ranking results；Wherein, information vector is not
Dependence and behavioral data, and can be carried out semantic extension, it can make the information recalled that there is diversity and novelty, label is avoided to call together
Bring result convergence problem is returned, to increase the diversity and novelty for recommending information, can more be felt for user development
The information of interest improves the clicking rate of information.
Detailed description of the invention
Fig. 1 is a kind of information recommended method flow chart provided in an embodiment of the present invention；
Fig. 2 is a kind of information recommended method flow chart provided in an embodiment of the present invention；
Fig. 3 is a kind of information recommended method flow chart provided in an embodiment of the present invention
Fig. 4 a is a kind of information recommended method flow chart provided in an embodiment of the present invention
Fig. 4 b is a kind of information recommended method flow chart provided in an embodiment of the present invention；
Fig. 5 is a kind of information recommendation apparatus structural block diagram provided in an embodiment of the present invention；
Fig. 6 is the structural schematic diagram of a kind of electronic equipment provided in an embodiment of the present invention.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limiting the invention.It also should be noted that in order to just
Only the parts related to the present invention are shown in description, attached drawing rather than entire infrastructure.
Fig. 1 is a kind of information recommended method flow chart provided in an embodiment of the present invention, and the method can be recommended by information
Device executes, and described device can realize that described device can be only fitted to terminal or server by software and/or hardware
In equal electronic equipments.
As shown in Figure 1, technical solution provided in an embodiment of the present invention includes:
S110: determining the M information that the user clicks recently, and by the arithmetic average of the label vector of each information
As corresponding information vector.
In embodiments of the present invention, the M information that user clicks recently refers to that the information for clicking user is suitable according to the time
Sequence is ranked up, M information for taking rear M information to click recently as user.Wherein, M is positive integer.
In an embodiment of the embodiment of the present invention, optionally, the label vector by each information counts
Average value is as corresponding information vector, comprising: is directed to each information, determines weighted value highest in all labels of the information
N label；The n label is separately input in term vector model, respectively each label vector；It is every based on what is obtained
A label vector calculates arithmetic average, and using the arithmetic average as information vector.Wherein, label can be to information
The identification of addition marks, and an information can have one or more label.Label vector can be by label be input to word to
Measure vector obtained in model, wherein the element in label vector can be numerical value.
Specifically, being directed to each information, all labels of information are ranked up from big to small according to weighted value, preceding n
Label is the highest n label of weighted value.Term vector model is trained model, and label is input to term vector model can be with
Obtain label vector.N label is separately input in term vector model, corresponding n label vector is obtained, calculates n mark
The arithmetic average for signing vector, using the arithmetic average as information vector.
For example, 4 information that user clicks recently, for one of information item, the tag set of item is should
Highest 4 labels of information weighted value, i.e. t1, t2, t3, t4.W2v_t { 1,4 } can indicate 200 dimensions of label t { 1,4 }
Word2vec (w2v) vector, item vector can indicate are as follows: emb_i, emb_i are that the arithmetic of 4 label word2vec vectors is flat
Mean value, it may be assumed that
Wherein, if label vector is not present in label, label vector is set as 200 dimension, 0 vector.
S120: the similarity based on the information vector of each information in the M information recalls K information.
In embodiments of the present invention, the calculation method of the similarity of information vector can be vector similarity in the related technology
Calculation method, such as the similarity of vector can be judged by entropy, or other methods in the related technology can also be passed through
Determine the similarity of vector.
It is optionally, described based on each information in the M information in an embodiment of the embodiment of the present invention
The similarity of information vector recalls K information, comprising: the money that clicking rate is unsatisfactory for the first preset condition is removed from all information
News, obtain remaining information；The remaining information is distinguished according to the information vector similarity of each information in the M information
It is ranked up, K1 information before taking respectively, and is summarized to obtain the K information recalled.
Wherein, the first preset condition, which can according to need, is set, for example, the first preset condition can be lower than overall
Clicking rate 50%.Wherein, overall clicking rate is the average value of the clicking rate of all information.
Specifically, counting the clicking rate of each information, the clicking rate and the overall of all information for counting each information are clicked
Rate, removal clicking rate are lower than the low quality information of overall clicking rate 50%.Calculate each information and corresponding information vector simultaneously
It saves into faiss (a kind of the dense vector index of open source and cluster frame).If what user clicked recently is 4 information,
The information vector that 4 information are respectively adopted recalls 16 highest information of similarity in faiss.
Information is recalled by using the similarity of information vector as a result, the extension of label semantically can be brought, avoid marking
Label recall bring result convergence problem.For example, recalling, recalling if carrying out information by label if label is England Premier League, Division A League Matches of Spanish Football
As a result it is difficult the result for Division A League Football Matches of Italy label occur in.
S130: the linked character of each information recalled and each information recalled is input to the mathematical modulo trained
In type, the clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate.
In embodiments of the present invention, mathematical model can be study order models, and study order models can use depth
Factorization machine (deep Factorization Machine, DeepFM) model, DeepFM model are a kind of while training FM
The depth model of model and DNN；Compared to FM model, DeepFM model not only has FM model to the energy of second order feature compositional modeling
Power, and combined using DNN model learning high-order feature, it is easier to recommend novel information out.
In embodiments of the present invention, optionally, linked character includes: clicking recently with user respectively for the information recalled
The information vector similarity of M information；Classification clicking rate/the hits for the information recalled；The click of label in the information recalled
Rate/click data；And the information recalled is directed to the characteristic of user.Optionally, the information recalled is directed to the feature of user
Data may include number, the behavioral data of each user etc. that each user clicks information.
Wherein, the information vector similarity for the M information of the information recalled clicked recently with user respectively can be recognized
To be information vector similarity feature, the exploring ability of user interest can be embodied, it can be with by the output result of data model
Embody the depth interest digging and exploring ability of user.Wherein, the classification clicking rate/hits for the information recalled, and recall
Information in label clicking rate/click data be considered session week in feedback characteristic, user interest can be embodied
Variation, the nearest interests change of user is reacted in recommendation results, body can be strengthened according to the output result of data model
The interest Real-time Feedback at current family, such as the long-term preference profiles of user are hobby footballs, but the 3 football classes money exposed recently
News are not clicked, then need for the negative-feedback information to be input in mathematical model, by by the feedback characteristic in session period
It is input in data model, the interest Real-time Feedback of user can be strengthened.
In embodiments of the present invention, it is input to by the linked character of each information recalled and each information recalled
It can also include being trained to mathematical model before in the mathematical model trained.Specifically: recommended information is directed to user
Characteristic, such as user behavior data, user's shot and long term characteristic etc. generates training sample；Wherein, short in order to embody
The feedback characteristic in the session period is added in phase interests change；Information vector similarity is added in order to embody the ability of interest exploration
Feature, can be to avoid the excessive use of model.Feedback characteristic in session week: the clicking rate of the exposure classification of recommended information/
Clicking rate/hits of hits, recommended information exposure label.Wherein, information vector similarity feature: recommended information
The similarity of information vector and the nearest 4 clicks information vector of user.Optionally, according to the above-mentioned various dimensions of recommended information
In feature inputting mathematical model, to be trained to model, and Ordering and marking is provided by model.
S140: information recommendation is carried out to user based on ranking results.
In embodiments of the present invention in an embodiment of embodiment, optionally, the information recalled can be based on point
It hits rate to be ranked up from big to small, the information of preset quantity can be taken from front to back, as information recommended to the user.
In an embodiment of the embodiment of the present invention, optionally, the information recalled can be based on to clicking rate from big
It is ranked up to small, and the information after sequence is based on setting restrictive condition and is reordered, take preset quantity from front to back
Information.The quantity that setting restrictive condition can be information generic or with theme is less than setting quantity or same calls together
The quantity for returning the information of strategy is less than setting quantity, or also can be set as needed other conditions.As a result, by being based on setting
Definite limitation condition reorders, and the diversity and novelty for recommending information can be improved, and the exposure of cold start-up information can be improved
Light ratio, allows system to recycle benign development.
Technical solution provided in an embodiment of the present invention passes through the mark of each information in M information clicking user recently
The arithmetic mean of instantaneous value of vector is signed as corresponding information vector, and the similarity based on information vector recalls K information, and will call together
It the information returned and recalls the linked character of each information and is input in mathematical model, the click for each information recalled
Rate, and be ranked up information based on clicking rate, information recommendation is carried out to user based on ranking results；Wherein, information vector is not
Dependence and behavioral data, and can be carried out semantic extension, avoid label from recalling bring result convergence problem, so as to be user
More interested information are provided, to improve the clicking rate of information.
Fig. 2 is a kind of information recommended method flow chart provided in an embodiment of the present invention, and the optinal plan in the present embodiment can
To be combined with one or more optinal plan in above-described embodiment.
In embodiments of the present invention, optionally, the method can also include:
Determine the preference label of user；
For each preference label of the user, multiple new information containing the preference label are determined, and from described more
N number of new information containing the preference label is recalled by upper-limit information circle UCB strategy in a new information.
As shown in Fig. 2, technical solution provided in an embodiment of the present invention includes:
S210: determining the M information that the user clicks recently, and by the arithmetic average of the label vector of each information
As corresponding information vector.
S220: the similarity based on the information vector of each information in the M information recalls K information.
S230: the preference label of user is determined.
In an embodiment in embodiments of the present invention, optionally, the preference label of the determining user, comprising:
It determines the information that user clicks in preset time period and the M1 information clicked recently, determines user to label in the information
Preference value；For the information that user in preset time period clicks, the label in information is carried out from big to small according to preference value
Sequence, and take preceding N1 label as the long-term preference label of user；For the M1 information that user clicks recently, according to preference value
The label in information is ranked up from big to small, N2 label is as the short-term preference label of user before taking.
Wherein, preset time period can be in nearest 1 month.M1 and N1 can be positive integer, and M1 can take 64, N1 can also
To take 64.Optionally, user can be following calculation to the preference value of label in information: for label t, u pairs of user
The preference value of label t is equal to the summation of the label t weighted value in the information set of the t containing label clicked, divided by label t's
Click total number of persons.Wherein, the information set of the t containing label of click refers to the information of the t containing label in the information of each user's click
The set of formation, label t has weighted value in each information, and weighted value of the label t in different information can be not
It is identical.
Specifically, for the information that user in nearest 1 month clicks, according to preference value from big to small to the label in information
It is ranked up, the first 64 long-term preference labels as user can be taken, for 64 information that user clicks recently, according to inclined
Good value is from big to small ranked up the label in information, short-term preference label of the N2 label as user before taking.Wherein, it uses
Head of a household's phase preference label and short-term preference label can be used as the preference label of user.Wherein, N2 is positive integer.
S240: for each preference label of the user, determining multiple new information containing the preference label, and from institute
It states in multiple new information and N number of new information containing the preference label is recalled by upper-limit information circle UCB strategy.
In embodiments of the present invention, specifically, being directed to each preference label, multiple new information of the label containing preference are determined,
Multiple new information are ranked up from high to low according to UCB score, take the new information of N number of label containing preference.Different preference marks
The quantity of the new information for the label containing preference that label are recalled can be different.
Specifically, 128 for can searching the t containing label by retrieval newly provide for example, being directed to user preference label t
News.When recalling the new information of the t containing the label for user u, need to calculate the UCB score of 128 new information, according to score height
Select several new information.
If it is 128 that the call back number of new information, which is sum, the call back number of the new information of label t be it is N number of, N is to recall
Sum is directed to the weight proportion of user multiplied by label t, it may be assumed thatWherein, biasu_tFor
Label t is directed to user's weighted value.
In embodiments of the present invention, optionally, it can be by the new information mode of recalling of UCB strategy such as under type: based on
Calculate the total exposure of 128 new information: total_pv=sum (pv_i)；Wherein, pv_i is the exposure rate of each new information,
Total_pv is total exposure rate.
Wherein it is possible to calculate each new information i UCB score based on following formula:
Wherein, clickiFor new information i
Clicking rate；UCB_i is the UCB score of i-th of new information；The weight accounting for utilizing part is explored in alpha parameter control, can use 0.4.
In embodiments of the present invention, for each preference label of user, by calculating the new money of each of label containing preference
The UCB score of news, and by new information according to being ranked up from big to small, takes the new information of top n label containing preference, as being directed to
The information that each preference label is recalled.
S250: the linked character of each information recalled and each information recalled is input to the mathematical modulo trained
In type, the clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate.
In embodiments of the present invention, the information recalled includes the K information that the similarity based on information vector is recalled, and
The new information recalled based on UCB strategy.
S260: information recommendation is carried out to user based on ranking results.
Wherein, the introduction of correlation step can refer to above-described embodiment in the embodiment of the present invention.
It should be noted that each step of S210-S240 executes sequence in technical solution provided in an embodiment of the present invention
It can be adjusted as required.It, can be before S210 perhaps S220 or S230 can be for example, S230 and S240
Before S210 or S220.
In the related art, information miscarriage product need user during constantly interacting with computer equipment, need
The novelty and diversity for recommending information are increased, the interest for better meeting user explores demand, that is, is meeting customer consumption
While relevent information, the interested information of user can also be expanded；The embodiment of the present invention is a by the M for clicking user recently
The arithmetic mean of instantaneous value of the label vector of each information is as corresponding information vector in information, and the similarity based on information vector
K information is recalled, information vector does not depend on and behavioral data, and can be carried out semantic extension, can make to recall information with multiplicity
Property, it avoids label from recalling bring result convergence problem, so as to provide more interested information for user, increases and recommend
The diversity and novelty of information.
In the related technology, information miscarriage product need the real-time interest of quick response user feedback, especially user to become
Change, avoids influencing information and recall, sort and recommend；And need to test out the quality of new information with the smallest flow cost
Quality reduces and explores cost；The logical preference label in conjunction with user of the embodiment of the present invention, calls new information together based on UCB strategy
It returns, the influence recalled, sort, recommended to information can be avoided with the real-time interests change of rapid feedback user, can accelerate newly to provide
The efficiency explored is interrogated, new information is significantly improved and explores the probability of exposure, and improve the click probability of new information.
Fig. 3 is a kind of information recommended method flow chart provided in an embodiment of the present invention, and the optinal plan in the present embodiment can
To be combined with one or more optinal plan in above-described embodiment.In embodiments of the present invention, optionally, the side
Method can also include:
From all information, recall containing the preference label, and overhead information parameters meet the information of the second preset condition.
As shown in figure 3, technical solution provided in an embodiment of the present invention includes:
S310: determining the M information that the user clicks recently, and by the arithmetic average of the label vector of each information
As corresponding information vector.
S320: the similarity based on the information vector of each information in the M information recalls K information.
S330: the preference label of user is determined.
S340: for each preference label of the user, determining multiple new information containing the preference label, and from institute
It states in multiple new information and N number of new information containing the preference label is recalled by upper-limit information circle UCB strategy.
S350: it from all information, recalls containing the preference label, and overhead information parameters meet the money of the second preset condition
News.
In embodiments of the present invention, overhead information parameters can be the temperature of information, the clicking rate etc. of information.Second preset condition
It can be temperature and be greater than the clicking rate of setting temperature perhaps information and be greater than setting clicking rate or can also be other conditions.
As a result, by recalling the preference label containing user from all information, and overhead information parameters meet the second preset condition
Information, the information of high quality can be recalled, by user click information recall other collaboration information, recalled by preference label
Generic popular information etc..
S360: the linked character of each information recalled and each information recalled is input to the mathematical modulo trained
In type, the clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate.
In embodiments of the present invention, the information recalled includes that the information recalled includes that the similarity based on information vector is recalled
K information, based on the new information that UCB strategy is recalled, and recall containing the preference label, and overhead information parameters meet second
The information of preset condition.
S370: information recommendation is carried out to user based on ranking results.
The introduction of correlation step can refer to the above embodiments in the embodiment of the present invention.
It should be noted that each step of S310-S350 executes sequence in technical solution provided in an embodiment of the present invention
It is not limited to above-mentioned sequence, can be adjusted as required.For example, S330 and S340, it can be in S310 or S320
Before or S330 can be before S310 or S320, and S350 can be executed after S330.
Fig. 4 a is a kind of information recommended method flow chart provided in an embodiment of the present invention, and the optinal plan in the present embodiment can
To be combined with one or more optinal plan in above-described embodiment.In embodiments of the present invention, optionally, the base
Information recommendation is carried out to user in ranking results, comprising:
Information after sequence is based on setting restrictive condition to reorder, information is carried out to user based on result is reordered
Recommend.
As shown in fig. 4 a, technical solution provided in an embodiment of the present invention includes:
S410: determining the M information that the user clicks recently, and by the arithmetic average of the label vector of each information
As corresponding information vector.
S420: the similarity based on the information vector of each information in the M information recalls K information.
S430: the preference label of user is determined；
S440: for each preference label of the user, determining multiple new information containing the preference label, and from institute
It states in multiple new information and N number of new information containing the preference label is recalled by upper-limit information circle UCB strategy.
S450: the linked character of each information recalled and each information recalled is input to the mathematical modulo trained
In type, the clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate.
S460: by the information after sequence be based on setting restrictive condition reorder, based on reorder result to user into
Row information is recommended.
It is ranked up from big to small specifically, the information recalled can be based on clicking rate, and by the information base after sequence
It reorders in setting restrictive condition, takes the information of preset quantity from front to back.Setting restrictive condition can be it is generic or
Person is less than setting quantity with the quantity that the quantity of the information of theme is less than setting quantity or the same information for recalling strategy,
Or it also can be set as needed other conditions.As a result, by being reordered based on setting restrictive condition, it can be improved and push away
The exposure ratio of cold start-up information can be improved in the diversity and novelty for recommending information.Specific side provided in an embodiment of the present invention
Method can also refer to Fig. 4 b.
Fig. 5 is a kind of information recommendation apparatus structural block diagram provided in an embodiment of the present invention, as shown in figure 5, described device packet
It includes information vector determining module 510, recall module 520, clicking rate determining module 530 and recommending module 540.
Wherein, information vector determining module 510, the M information clicked recently for determining user, and by each information
The arithmetic average of label vector is as corresponding information vector；
Module 520 is recalled, recalls K money for the similarity based on the information vector of each information in the M information
News；
Clicking rate determining module 530, for the linked character of each information recalled and each information recalled is defeated
Enter into the mathematical model trained, the clicking rate for each information recalled；
Recommending module 540 for being ranked up based on the clicking rate to all information recalled, and is based on ranking results
Information recommendation is carried out to user.
Optionally, information vector determining module 510, is used for:
For each information, the highest n label of weighted value in all labels of the information is determined；
The n label is separately input in term vector model, respectively each label vector；
Arithmetic average is calculated based on obtained each label vector, and using the arithmetic average as information vector.
Optionally, module 520 is recalled, is used for:
The information that clicking rate is unsatisfactory for the first preset condition is removed from all information, obtains remaining information；
The remaining information is arranged respectively according to the information vector similarity with each information in the M information
Sequence, K1 information before taking respectively, and summarized to obtain the K information recalled.
Optionally, module 520 is recalled, is also used to the association of each information recalled and each information recalled is special
Before sign is input in the mathematical model trained, the preference label of user is determined；
For each preference label of the user, multiple new information containing the preference label are determined, and from described more
N number of new information containing the preference label is recalled by upper-limit information circle UCB strategy in a new information.
Optionally, module 520 is recalled, is also used to the association of each information recalled and each information recalled is special
Before sign is input in the mathematical model trained, from all information, recall containing the preference label, and overhead information parameters meet
The information of second preset condition.
Optionally, recommending module 540 are reordered for the information after sequence to be based on setting restrictive condition, are based on
The result that reorders carries out information recommendation to user.
Optionally, the preference label of the determining user, comprising:
It determines the information that user clicks in preset time period and the M1 information clicked recently, determines user to the money
The preference value of label in news；
For the information that user in preset time period clicks, the label in information is arranged from big to small according to preference value
Sequence, and take preceding N1 label as the long-term preference label of user；
For the M1 information that user clicks recently, the label in information is ranked up from big to small according to preference value,
N2 label is as the short-term preference label of user before taking.
Optionally, linked character includes: the information vector for the M information of the information recalled clicked recently with user respectively
Similarity；Classification clicking rate/the hits for the information recalled；Clicking rate/click data of label in the information recalled；And it calls together
The information returned is directed to the characteristic of user；
The mathematical model is depth factor disassembler DeepFM model.
Method provided by any embodiment of the invention can be performed in above-mentioned apparatus, has the corresponding functional module of execution method
And beneficial effect.
Fig. 6 is a kind of device structure schematic diagram provided in an embodiment of the present invention, as shown in fig. 6, the equipment includes:
One or more processors 610, in Fig. 6 by taking a processor 610 as an example；
Memory 620；
The equipment can also include: input unit 630 and output device 640.
Processor 610, memory 620, input unit 630 and output device 640 in the equipment can pass through bus
Or other modes connect, in Fig. 6 for being connected by bus.
Memory 620 be used as a kind of non-transient computer readable storage medium, can be used for storing software program, computer can
Execute program and module, as the corresponding program instruction/module of one of embodiment of the present invention information recommended method (for example,
Attached information vector determining module 510 shown in fig. 5 recalls module 520, clicking rate determining module 530 and recommending module 540).Place
Software program, instruction and the module that reason device 610 is stored in memory 620 by operation, thereby executing computer equipment
A kind of information recommended method of above method embodiment is realized in various function application and data processing, it may be assumed that
Determine the M information that user clicks recently, and using the arithmetic average of the label vector of each information as correspondence
Information vector；
Similarity based on the information vector of each information in the M information recalls K information；
The linked character of each information recalled and each information recalled is input in the mathematical model trained,
The clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate；
Information recommendation is carried out to user based on ranking results.
Memory 620 may include storing program area and storage data area, wherein storing program area can store operation system
Application program required for system, at least one function；Storage data area can be stored to be created according to using for computer equipment
Data etc..In addition, memory 620 may include high-speed random access memory, it can also include non-transitory memory, such as
At least one disk memory, flush memory device or other non-transitory solid-state memories.In some embodiments, it stores
Optional device 620 includes the memory remotely located relative to processor 610, these remote memories can be by being connected to the network extremely
Terminal device.The example of above-mentioned network includes but is not limited to internet, intranet, local area network, mobile radio communication and its group
It closes.
Input unit 630 can be used for receiving the number or character information of input, and generate the user with computer equipment
Setting and the related key signals input of function control.Output device 640 may include that display screen etc. shows equipment.
The embodiment of the invention provides a kind of computer readable storage mediums, are stored thereon with computer program, the program
A kind of such as information recommended method provided in an embodiment of the present invention is realized when being executed by processor:
Determine the M information that user clicks recently, and using the arithmetic average of the label vector of each information as correspondence
Information vector；
Similarity based on the information vector of each information in the M information recalls K information；
The linked character of each information recalled and each information recalled is input in the mathematical model trained,
The clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate；
Information recommendation is carried out to user based on ranking results.
It can be using any combination of one or more computer-readable media.Computer-readable medium can be calculating
Machine readable signal medium or computer readable storage medium.Computer readable storage medium for example can be --- but it is unlimited
In system, device or the device of --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, or any above combination.It calculates
The more specific example (non exhaustive list) of machine readable storage medium storing program for executing includes: electrical connection with one or more conducting wires, just
Taking formula computer disk, hard disk, random access memory (RAM), read-only memory (ROM), erasable type may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In this document, computer readable storage medium can be it is any include or storage journey
The tangible medium of sequence, the program can be commanded execution system, device or device use or in connection.
Computer-readable signal media may include in a base band or as carrier wave a part propagate data-signal,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including --- but
It is not limited to --- electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be
Any computer-readable medium other than computer readable storage medium, which can send, propagate or
Transmission is for by the use of instruction execution system, device or device or program in connection.
The program code for including on computer-readable medium can transmit with any suitable medium, including --- but it is unlimited
In --- wireless, electric wire, optical cable, RF etc. or above-mentioned any appropriate combination.
The computer for executing operation of the present invention can be write with one or more programming languages or combinations thereof
Program code, described program design language include object oriented program language-such as Java, Smalltalk, C++,
Further include conventional procedural programming language-such as " C " language or similar programming language.Program code can be with
It fully executes, partly execute on the user computer on the user computer, being executed as an independent software package, portion
Divide and partially executes or executed on a remote computer or server completely on the remote computer on the user computer.In
Be related in the situation of remote computer, remote computer can pass through the network of any kind --- including local area network (LAN) or
Wide area network (WAN)-be connected to subscriber computer, or, it may be connected to outer computer (such as mentioned using Internet service
It is connected for quotient by internet).
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The invention is not limited to the specific embodiments described herein, be able to carry out for a person skilled in the art it is various it is apparent variation,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above embodiments to the present invention
It is described in further detail, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, also
It may include more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (11)
1. a kind of information recommended method characterized by comprising
Determine the M information that user clicks recently, and using the arithmetic average of the label vector of each information as corresponding money
Interrogate vector；
Similarity based on the information vector of each information in the M information recalls K information；
The linked character of each information recalled and each information recalled is input in the mathematical model trained, is obtained
The clicking rate for each information recalled, and all information recalled are ranked up based on the clicking rate；
Information recommendation is carried out to user based on ranking results.
2. the method according to claim 1, wherein the arithmetic average of the label vector by each information
As corresponding information vector, comprising:
For each information, the highest n label of weighted value in all labels of the information is determined；
The n label is separately input in term vector model, respectively each label vector；
Arithmetic average is calculated based on obtained each label vector, and using the arithmetic average as information vector.
3. the method according to claim 1, wherein the information based on each information in the M information
The similarity of vector recalls K information, comprising:
The information that clicking rate is unsatisfactory for the first preset condition is removed from all information, obtains remaining information；
The remaining information is ranked up respectively according to the information vector similarity with each information in the M information, point
K1 information before not taking, and summarized to obtain the K information recalled.
4. the method according to claim 1, wherein by each information recalled and each information recalled
Linked character be input in the mathematical model trained before, further includes:
Determine the preference label of user；
For each preference label of the user, multiple new information containing the preference label are determined, and from the multiple new
N number of new information containing the preference label is recalled by upper-limit information circle UCB strategy in information.
5. according to the method described in claim 4, it is characterized in that, by each information recalled and each information recalled
Linked character be input in the mathematical model trained before, further includes:
From all information, recall containing the preference label, and overhead information parameters meet the information of the second preset condition.
6. the method according to claim 1, wherein it is described based on ranking results to user carry out information recommendation,
Include:
Information after sequence is based on setting restrictive condition to reorder, information is carried out to user based on the result that reorders and is pushed away
It recommends.
7. according to the method described in claim 2, it is characterized in that, the preference label of the determining user, comprising:
It determines the information that user clicks in preset time period and the M1 information clicked recently, determines user in the information
The preference value of label；
For the information that user in preset time period clicks, the label in information is ranked up from big to small according to preference value,
And take preceding N1 label as the long-term preference label of user；
For the M1 information that user clicks recently, the label in information is ranked up from big to small according to preference value, before taking
N2 label is as the short-term preference label of user.
8. the method according to claim 1, wherein
Linked character includes: the information vector similarity for the M information of the information recalled clicked recently with user respectively；It recalls
Information classification clicking rate/hits；Clicking rate/click data of label in the information recalled；And the information needle recalled
To the characteristic of user；
The mathematical model is depth factor disassembler DeepFM model.
9. a kind of information recommendation apparatus characterized by comprising
Information vector determining module, the M information clicked recently for determining user, and by the calculation of the label vector of each information
Number average value is as corresponding information vector；
Module is recalled, recalls K information for the similarity based on the information vector of each information in the M information；
Clicking rate determining module has been instructed for the linked character of each information recalled and each information recalled to be input to
In experienced mathematical model, the clicking rate for each information recalled；
Recommending module, for being ranked up based on the clicking rate to all information recalled, and based on ranking results to user
Carry out information recommendation.
10. a kind of electronic equipment characterized by comprising
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
Now such as a kind of described in any item information recommended methods of claim 1-8.
11. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
A kind of such as claim 1-8 described in any item information recommended methods are realized when execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910712063.2A CN110457581B (en) | 2019-08-02 | 2019-08-02 | Information recommendation method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910712063.2A CN110457581B (en) | 2019-08-02 | 2019-08-02 | Information recommendation method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110457581A true CN110457581A (en) | 2019-11-15 |
| CN110457581B CN110457581B (en) | 2023-04-18 |
Family
ID=68484692
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910712063.2A Active CN110457581B (en) | 2019-08-02 | 2019-08-02 | Information recommendation method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110457581B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110930223A (en) * | 2019-11-21 | 2020-03-27 | 苏州朗动网络科技有限公司 | Recommendation recall method, device and storage medium based on field-aware factorization machine |
| CN111078997A (en) * | 2019-11-18 | 2020-04-28 | 中国银行股份有限公司 | Information recommendation method and device |
| CN111159341A (en) * | 2019-12-25 | 2020-05-15 | 中国建设银行股份有限公司 | Information recommendation method and device based on user investment and financing preference |
| CN111191136A (en) * | 2019-12-30 | 2020-05-22 | 华为技术有限公司 | Information recommendation method and related equipment |
| CN111310058A (en) * | 2020-03-27 | 2020-06-19 | 北京百度网讯科技有限公司 | Information theme recommendation method and device, terminal and storage medium |
| CN111324813A (en) * | 2020-02-20 | 2020-06-23 | 深圳前海微众银行股份有限公司 | Recommendation method, device, equipment and computer readable storage medium |
| CN111324733A (en) * | 2020-02-07 | 2020-06-23 | 北京创鑫旅程网络技术有限公司 | Content recommendation method, device, equipment and storage medium |
| CN111382361A (en) * | 2020-03-12 | 2020-07-07 | 腾讯科技（深圳）有限公司 | Information pushing method and device, storage medium and computer equipment |
| CN111444438A (en) * | 2020-03-24 | 2020-07-24 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining recall permission rate of recall strategy |
| CN111538902A (en) * | 2020-04-15 | 2020-08-14 | 达而观信息科技（上海）有限公司 | Information pushing method and device, storage medium and electronic equipment |
| CN111581545A (en) * | 2020-05-12 | 2020-08-25 | 腾讯科技（深圳）有限公司 | Method for sorting recalled documents and related equipment |
| CN111898028A (en) * | 2020-08-07 | 2020-11-06 | 北京小米移动软件有限公司 | Entity object recommendation method, device and storage medium |
| CN112328906A (en) * | 2020-11-04 | 2021-02-05 | 中国平安人寿保险股份有限公司 | Content item recommendation method, device, equipment and storage medium |
| CN112989118A (en) * | 2021-02-04 | 2021-06-18 | 北京奇艺世纪科技有限公司 | Video recall method and device |
| CN113450172A (en) * | 2020-03-27 | 2021-09-28 | 北京沃东天骏信息技术有限公司 | Commodity recommendation method and device |
| CN113656694A (en) * | 2021-08-18 | 2021-11-16 | 中国平安人寿保险股份有限公司 | Information recommendation method, device and equipment based on machine learning and storage medium |
| CN113688295A (en) * | 2021-10-26 | 2021-11-23 | 北京达佳互联信息技术有限公司 | Data determination method and device, electronic equipment and storage medium |
| CN114116982A (en) * | 2021-11-30 | 2022-03-01 | 中国建设银行股份有限公司 | Method, system, device, storage medium and program product for content distribution |
| CN114201652A (en) * | 2021-12-14 | 2022-03-18 | 宁波深擎信息科技有限公司 | Information recall ordering method, device and equipment based on label classification |
| CN115587261A (en) * | 2022-12-09 | 2023-01-10 | 思创数码科技股份有限公司 | Government affair resource catalog recommendation method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107844525A (en) * | 2017-10-12 | 2018-03-27 | 广州艾媒数聚信息咨询股份有限公司 | A kind of information personalized recommendation method, system and device based on user behavior |
| WO2018157818A1 (en) * | 2017-03-02 | 2018-09-07 | 广州市动景计算机科技有限公司 | Method and apparatus for inferring preference of user, terminal device, and storage medium |
| JP2018147045A (en) * | 2017-03-01 | 2018-09-20 | ヤフー株式会社 | Retrieval device, retrieval method, and retrieval program |
-
2019
- 2019-08-02 CN CN201910712063.2A patent/CN110457581B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2018147045A (en) * | 2017-03-01 | 2018-09-20 | ヤフー株式会社 | Retrieval device, retrieval method, and retrieval program |
| WO2018157818A1 (en) * | 2017-03-02 | 2018-09-07 | 广州市动景计算机科技有限公司 | Method and apparatus for inferring preference of user, terminal device, and storage medium |
| CN107844525A (en) * | 2017-10-12 | 2018-03-27 | 广州艾媒数聚信息咨询股份有限公司 | A kind of information personalized recommendation method, system and device based on user behavior |
Non-Patent Citations (2)
| Title |
|---|
| 磨莉: "深度神经网络在点击率预测中应用研究", 《科技通报》 * |
| 陈学辉等: "基于内容推荐的资讯推荐系统的设计与实现", 《电脑知识与技术》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111078997A (en) * | 2019-11-18 | 2020-04-28 | 中国银行股份有限公司 | Information recommendation method and device |
| CN111078997B (en) * | 2019-11-18 | 2023-08-22 | 中国银行股份有限公司 | Information recommendation method and device |
| CN110930223A (en) * | 2019-11-21 | 2020-03-27 | 苏州朗动网络科技有限公司 | Recommendation recall method, device and storage medium based on field-aware factorization machine |
| CN111159341A (en) * | 2019-12-25 | 2020-05-15 | 中国建设银行股份有限公司 | Information recommendation method and device based on user investment and financing preference |
| CN111159341B (en) * | 2019-12-25 | 2023-04-14 | 中国建设银行股份有限公司 | Information recommendation method and device based on user investment and financial management preference |
| CN111191136A (en) * | 2019-12-30 | 2020-05-22 | 华为技术有限公司 | Information recommendation method and related equipment |
| CN111324733A (en) * | 2020-02-07 | 2020-06-23 | 北京创鑫旅程网络技术有限公司 | Content recommendation method, device, equipment and storage medium |
| CN111324813A (en) * | 2020-02-20 | 2020-06-23 | 深圳前海微众银行股份有限公司 | Recommendation method, device, equipment and computer readable storage medium |
| CN111382361A (en) * | 2020-03-12 | 2020-07-07 | 腾讯科技（深圳）有限公司 | Information pushing method and device, storage medium and computer equipment |
| CN111382361B (en) * | 2020-03-12 | 2023-05-02 | 腾讯科技（深圳）有限公司 | Information pushing method, device, storage medium and computer equipment |
| CN111444438A (en) * | 2020-03-24 | 2020-07-24 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining recall permission rate of recall strategy |
| CN111444438B (en) * | 2020-03-24 | 2023-09-01 | 北京百度网讯科技有限公司 | Method, device, equipment and storage medium for determining quasi-recall rate of recall strategy |
| CN111310058B (en) * | 2020-03-27 | 2023-08-08 | 北京百度网讯科技有限公司 | Information theme recommendation method, device, terminal and storage medium |
| CN113450172A (en) * | 2020-03-27 | 2021-09-28 | 北京沃东天骏信息技术有限公司 | Commodity recommendation method and device |
| CN111310058A (en) * | 2020-03-27 | 2020-06-19 | 北京百度网讯科技有限公司 | Information theme recommendation method and device, terminal and storage medium |
| CN111538902A (en) * | 2020-04-15 | 2020-08-14 | 达而观信息科技（上海）有限公司 | Information pushing method and device, storage medium and electronic equipment |
| CN111538902B (en) * | 2020-04-15 | 2024-03-22 | 达观数据有限公司 | Information pushing method and device, storage medium and electronic equipment |
| CN111581545A (en) * | 2020-05-12 | 2020-08-25 | 腾讯科技（深圳）有限公司 | Method for sorting recalled documents and related equipment |
| CN111581545B (en) * | 2020-05-12 | 2023-09-19 | 腾讯科技（深圳）有限公司 | Method for sorting recall documents and related equipment |
| CN111898028B (en) * | 2020-08-07 | 2024-04-19 | 北京小米移动软件有限公司 | Entity object recommendation method, device and storage medium |
| CN111898028A (en) * | 2020-08-07 | 2020-11-06 | 北京小米移动软件有限公司 | Entity object recommendation method, device and storage medium |
| CN112328906B (en) * | 2020-11-04 | 2024-08-27 | 中国平安人寿保险股份有限公司 | Content item recommendation method, device, equipment and storage medium |
| CN112328906A (en) * | 2020-11-04 | 2021-02-05 | 中国平安人寿保险股份有限公司 | Content item recommendation method, device, equipment and storage medium |
| CN112989118B (en) * | 2021-02-04 | 2023-08-18 | 北京奇艺世纪科技有限公司 | Video recall method and device |
| CN112989118A (en) * | 2021-02-04 | 2021-06-18 | 北京奇艺世纪科技有限公司 | Video recall method and device |
| CN113656694B (en) * | 2021-08-18 | 2023-07-25 | 中国平安人寿保险股份有限公司 | Information recommendation method, device, equipment and storage medium based on machine learning |
| CN113656694A (en) * | 2021-08-18 | 2021-11-16 | 中国平安人寿保险股份有限公司 | Information recommendation method, device and equipment based on machine learning and storage medium |
| CN113688295A (en) * | 2021-10-26 | 2021-11-23 | 北京达佳互联信息技术有限公司 | Data determination method and device, electronic equipment and storage medium |
| CN114116982A (en) * | 2021-11-30 | 2022-03-01 | 中国建设银行股份有限公司 | Method, system, device, storage medium and program product for content distribution |
| CN114201652A (en) * | 2021-12-14 | 2022-03-18 | 宁波深擎信息科技有限公司 | Information recall ordering method, device and equipment based on label classification |
| CN114201652B (en) * | 2021-12-14 | 2024-07-16 | 宁波深擎信息科技有限公司 | Information recall ordering method, device and equipment based on label grading |
| CN115587261A (en) * | 2022-12-09 | 2023-01-10 | 思创数码科技股份有限公司 | Government affair resource catalog recommendation method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110457581B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110457581A (en) | A kind of information recommended method, device, electronic equipment and storage medium | |
| CN108304379B (en) | Article identification method and device and storage medium | |
| US10949000B2 (en) | Sticker recommendation method and apparatus | |
| CN108776676B (en) | Information recommendation method and device, computer readable medium and electronic device | |
| CN110543598B (en) | Information recommendation method and device and terminal | |
| CN110909182B (en) | Multimedia resource searching method, device, computer equipment and storage medium | |
| CN109684554A (en) | The determination method and news push method of the potential user of news | |
| US12020267B2 (en) | Method, apparatus, storage medium, and device for generating user profile | |
| CN111241394B (en) | Data processing method, data processing device, computer readable storage medium and electronic equipment | |
| CN109165974A (en) | A kind of commercial product recommending model training method, device, equipment and storage medium | |
| CN109168047B (en) | Video recommendation method and device, server and storage medium | |
| US10726087B2 (en) | Machine learning system and method to identify and connect like-minded users | |
| CN105718566A (en) | Intelligent music recommendation system | |
| CN116894711A (en) | Commodity recommendation reason generation method and device and electronic equipment | |
| CN110472016A (en) | Article recommended method, device, electronic equipment and storage medium | |
| CN113657087A (en) | Information matching method and device | |
| CN117391824B (en) | Method and device for recommending articles based on large language model and search engine | |
| CN112948602B (en) | Content display method, device, system, equipment and storage medium | |
| CN112115354B (en) | Information processing method, device, server and storage medium | |
| CN110442789A (en) | Method, apparatus and electronic equipment are determined based on the association results of user behavior | |
| CN108694171B (en) | Information pushing method and device | |
| CN113704596B (en) | Method and apparatus for generating recall information sets | |
| CN114090848A (en) | Data recommendation and classification method, feature fusion model and electronic equipment | |
| CN114328992A (en) | Multimedia information recommendation method, device, program product, equipment and medium | |
| CN113744011A (en) | Article collocation method and article collocation device |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |