CN106952111A - Personalized recommendation method and device - Google Patents

Personalized recommendation method and device Download PDF

Info

Publication number
CN106952111A
CN106952111A CN201710108664.3A CN201710108664A CN106952111A CN 106952111 A CN106952111 A CN 106952111A CN 201710108664 A CN201710108664 A CN 201710108664A CN 106952111 A CN106952111 A CN 106952111A
Authority
CN
China
Prior art keywords
article
behavior
destination object
recommended
similarity
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
Application number
CN201710108664.3A
Other languages
Chinese (zh)
Other versions
CN106952111B (en
Inventor
李骏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Neusoft Corp
Original Assignee
Neusoft Corp
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Neusoft Corp filed Critical Neusoft Corp
Priority to CN201710108664.3A priority Critical patent/CN106952111B/en
Publication of CN106952111A publication Critical patent/CN106952111A/en
Application granted granted Critical
Publication of CN106952111B publication Critical patent/CN106952111B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
    • G06Q30/0224—Discounts or incentives, e.g. coupons or rebates based on user history

Abstract

The present invention provides a kind of personalized recommendation method and device, the technical problem relatively low to solve particulate degree of the prior art for the judgement of article similarity.This method includes：Identical article among the first article set of destination object and the second article set of the first object is determined, first object is any object in object set；Determine the behavior vector of the destination object and first object respectively for the article；According to the behavior is vectorial, the first article set and the second article set calculate similarity between the destination object and first object；Similarity between each object in the destination object and the object set carries out article recommendation.

Description

Personalized recommendation method and device
Technical field
The present invention relates to field of information processing, in particular it relates to a kind of personalized recommendation method and device.
Background technology
Personalized recommendation system has been widely used for various fields now, and it can find similar with targeted customer's interest User's set, and from the user gather browsed article in for targeted customer recommend article.Or, personalized recommendation system The article set similar to the target item of targeted customer can be found, and recommends article from article set for targeted customer.
Existing personalized recommendation system generally uses UserCF (User Collaboration Filter, based on user's Collaborative filtering) and ItemCF (Item Collaboration Filter, the collaborative filtering based on article), two kinds Calculating in algorithm for similarity is generally Jie Kade Jaccard similarities or cosine similarity.
Wherein, the computing formula of Jaccard similarities is as follows：
W=| N (u) ∩ N (v) |/| N (u) ∪ N (v) |；
The computing formula of cosine similarity is as follows：
Wherein, w represents similarity, N (u), and N (v) represents that user u, v had the article set of behavior respectively.
From above-mentioned computing formula, the calculating for two kinds of similarities that prior art is used all only accounts for user u and use Family v had the article of behavior in itself, influence of the other factors to similarity was not considered, that is to say, that prior art is for article The particulate degree of the judgement of similarity is relatively low, cause user between the tendency interested of different articles and user to phase jljl Difference between the tendency interested of product embodies not enough.
The content of the invention
The main object of the present invention is to provide a kind of personalized recommendation method and device, to solve prior art for thing The relatively low technical problem of the particulate degree of the judgement of product similarity.
To achieve these goals, the present invention provides a kind of personalized recommendation method, including：
Identical article among the first article set of destination object and the second article set of the first object is determined, it is described First object is any object in object set；
Determine the behavior vector of the destination object and first object respectively for the article；
According to the behavior is vectorial, the first article set and the second article set calculate the destination object with Similarity between first object；
Similarity between each object in the destination object and the object set carries out article recommendation.
Alternatively, it is described according to the behavior is vectorial, the first article set and the second article set calculate institute The similarity stated between destination object and first object includes：
Similarity between the destination object and first object is calculated according to equation below：
Wherein, N (u) represents the first article set of the destination object u, and N (v) represents the first object v's The second article set, Ωu∩vIt is identical article among the first article set and the second article set, It is behavior vector of the destination object for article t,It is behavior vector of first object for article t, w is institute State similarity.
Alternatively, a kind of behavior of each element correspondence in the behavior vector, the value of each element is corresponding behavior Weights, wherein, the weights are that the comment content for the article is analyzed by natural language processing NLP technologies Obtained comment content similarities either emotion uniformity, or, the weights are the confidence levels of comment behavior, or, institute It is the numerical value obtained based on term frequency-inverse document frequency TF-IDF statistics to state weights.
Alternatively, the similarity between each object in the destination object and the object set is carried out Article is recommended to include：
Article to be recommended is determined from the article set of other objects similar to the destination object；
Calculate interest level of the destination object to each article to be recommended；
Recommendation sequence is carried out to each article to be recommended according to the interest level.
Alternatively, the interest level for calculating the destination object to each article to be recommended, including：
Interest level of the destination object to each article to be recommended is calculated by equation below：
Wherein, p*(u, i) represents interest levels of the destination object u to article i, and S (u, K) is included and destination object u is emerging The immediate K object of interest, N (i) is the object set for having behavior to article i, wuvIt is the destination object u and described first Object v similarity, riIt is significance levels of the article i in the article to be recommended, nhIt is that the K object had to article h The number of times of behavior, Φ is the article to be recommended, NjIt is the number of times that the object set had behavior to article j, Ω is all Article.
The present invention also provides a kind of personalized recommendation device, including：
First determining unit, for determine the first article set of destination object and the first object the second article set it Middle identical article, first object is any object in object set；
Second determining unit, for determining that the destination object and first object are directed to the behavior of the article respectively Vector；
Computing unit, for according to the behavior is vectorial, the first article set and the second article set are calculated Similarity between the destination object and first object；
Article recommendation unit, for according to similar between the destination object and each object in the object set Degree carries out article recommendation.
Alternatively, second determining unit is used for：
Similarity between the destination object and first object is calculated according to equation below：
Wherein, N (u) represents the first article set of the destination object u, and N (v) represents the first object v's The second article set, Ωu∩vIt is identical article among the first article set and the second article set,It is The destination object is vectorial for article t behavior,It is behavior vector of first object for article t, w is described Similarity.
Alternatively, a kind of behavior of each element correspondence in the behavior vector, the value of each element is corresponding behavior Weights, wherein, the weights are that the comment content for the article is analyzed by natural language processing NLP technologies Obtained comment content similarities either emotion uniformity, or, the weights are the confidence levels of comment behavior, or, institute It is the numerical value obtained based on term frequency-inverse document frequency TF-IDF statistics to state weights.
Alternatively, the article recommendation unit is used for：
Article to be recommended is determined from the article set of other objects similar to the destination object；
Calculate interest level of the destination object to each article to be recommended；
Recommendation sequence is carried out to each article to be recommended according to the interest level.
Alternatively, the article recommendation unit is used for：
Interest level of the destination object to each article to be recommended is calculated by equation below：
Wherein, p*(u, i) represents interest levels of the destination object u to article i, and S (u, K) is included and destination object u is emerging The immediate K object of interest, N (i) is the object set for having behavior to article i, wuvIt is the destination object u and described first Object v similarity, riIt is significance levels of the article i in the article to be recommended, nhIt is that the K object had to article h The number of times of behavior, Φ is the article to be recommended, NjIt is the number of times that the object set had behavior to article j, Ω is all Article.
Above-mentioned technical proposal not only allows for thing common between destination object and the first object when calculating similarity Product, and the concrete behavior that destination object and the first object had for the article is considered, add the thin of similarity judgement Grain degree, has reacted the difference between user or article, and different user to the difference of same article interest level, so that It can recommend more to meet the article of demand for user, lift Consumer's Experience.
Other features and advantages of the present invention will be described in detail in subsequent embodiment part.
Brief description of the drawings
Accompanying drawing is, for providing a further understanding of the present invention, and to constitute a part for specification, with following tool Body embodiment is used to explain the present invention together, but is not construed as limiting the invention.In the accompanying drawings：
Fig. 1 is a kind of schematic flow sheet of personalized recommendation method provided in an embodiment of the present invention；
Fig. 2 is the schematic flow sheet that a kind of article provided in an embodiment of the present invention recommends sort method；
Fig. 3 is a kind of structural representation of personalized recommendation device provided in an embodiment of the present invention.
Embodiment
The embodiment of the present invention is described in detail below in conjunction with accompanying drawing.It should be appreciated that this place is retouched The embodiment stated is merely to illustrate and explain the present invention, and is not intended to limit the invention.
Fig. 1 provides a kind of schematic flow sheet of personalized recommendation method for the embodiment of the present invention, as illustrated, this method bag Include：
S101, the first article set for determining destination object and identical thing among the second article set of the first object Product, first object is any object in object set.
Wherein, destination object can be the targeted customer for needing to recommend for it article, then the first article set is the mesh Mark user had the set of the article of behavior.Wherein, behavior of the user to article refers to browse, comment, shares, and collects, thumb up Etc., for different application scene, user may be different for the behavior of article.Similarly, object set can refer to that target is used User's set of family concern, the first object refers to any user of targeted customer's concern, in this case, the embodiment of the present invention Described personalized recommendation method refers to recommend the possible thing interested of targeted customer at the other users paid close attention to from targeted customer Product.
In addition, what deserves to be explained is, personalized recommendation method provided in an embodiment of the present invention can apply to each application Scene, under different application scenarios, the specific things that article is referred to may be different, for example, under the scene of shopping on the web, Article refers to the commodity sold on the net, under social scene, and article recommends that friend recommendation can be referred to, can also refer to other The state of user's issue.
S102, determine the destination object and first object respectively for the article behavior vector.
For example, behavior of the user to article is generally comprised in news category commending system：Browse, comment on, share, collect, Thumb up.In this case, behavior vector can be expressed as a=(x1, x2, x3, x4, x5), wherein, behavior vector in each A kind of behavior of element correspondence, for example, x1Correspondence is browsed, x2Correspondence comment, x3Correspondence is shared, x4Correspondence collection, x5Correspondence thumb up.OK Value for each vectorial element can represent whether user carried out the behavior to article, for example, 0 represents not carry out, 1 represents Carry out.Illustratively,Represent that user v had to article t and browsed, comment on and share Behavior,Then represent that user u had comment to article t, shared the behavior with thumb up.
S103, according to the behavior is vectorial, the first article set and the second article set calculate the target Similarity between object and first object.
In a kind of possible implementation of the embodiment of the present invention, step S103 can specifically be calculated according to equation below Similarity between the destination object and first object：
Wherein, N (u) represents the first article set of the destination object u, and N (v) represents the first object v's The second article set, Ωu∩vIt is identical article among the first article set and the second article set, It is behavior vector of the destination object for article t,It is behavior vector of first object for article t, w is institute State similarity.
Illustratively, user's set U={ A, B, C }, article set T={ a, b, c, d, e, f, g }, behavior set Act=is { clear Looking at, comment on, share, collect, thumb up, user is to the behavior record of article：
A：A (is browsed, comment on), and b (is browsed, comment on, share), and c (is browsed, collect, thumb up)；
B：B (is browsed, comment on), and c (is browsed, comment on, share), and d (is browsed, shared, thumb up), f (browse, comment on, collection, Thumb up)；
C：B (is browsed, comment on, share), and c (is browsed, commented on, collection, thumb up), and e (is browsed, collect, thumb up), g (browse, Comment, shares, and collects, thumb up).
In this case, above-mentioned destination object can be user A, and its corresponding first article collection is combined into { a, b, c }, on It can be user B to state the first object, and its corresponding second article collection is combined into { b, c, d, f }.Then the first article set and the second thing The identical article that user A and user B had behavior in product set is article b and article c.Also, user A is to article b's Behavior vector is (1,1,1,0,0), and user B is (1,1,0,0,0), behaviors of the user A to article c to article b behavior vector Vector is (1,0,0,1,1), and user B is (1,1,1,0,0) to article c behavior vector.
Further, user A and user B similarity w are can be calculated according to above-mentioned formulaAB：
Similarly, user A and user C similarity w be can be calculatedAC：
It follows that the similarity between user A and user B is less than the similarity between user A and user C.
In order to more intuitively embody the technique effect that method provided in an embodiment of the present invention can reach, below to existing There is technology to obtain similarity for the example above calculating to illustrate：
In the prior art, the Jaccard similarities w between user A and user BAB, and between user A and user C Jaccard similarities wACIt is as follows respectively：
In addition, the cosine similarity w between user A and user BAB, and the cosine similarity between user A and user C wACIt is as follows respectively：
From the foregoing, it will be observed that prior art can not embody differences of the user A respectively between user B and user C, and the present invention is adopted Similarity calculating method, can preferably distinguish the similarity degree between user.
What deserves to be explained is, above-mentioned to be merely illustrative, the value of each element in the behavior vector indicates user Whether corresponding behavior is carried out to article.
In the alternatively possible implementation of the embodiment of the present invention, a kind of row of each element correspondence in behavior vector For, the value of each element is the weights of corresponding behavior, wherein, the weights are by NLP (Neural Language Processing, natural language processing) technology to for the article comment content carry out analyze obtained comment content phase Like property either emotion uniformity, or, the weights are the confidence levels of comment behavior, or, the weights are to be based on TF- IDF (term frequency-inverse document frequency, term frequency-inverse document frequency) statistics is obtained Numerical value.
That is, the value of each element in behavior vector not only shows whether user carries out corresponding row to article Also to show that the weight of other behaviors is compared in this behavior.For example, for user for the comment behavior that article is carried out, no Same article may be commented on user, in order to symbolize otherness, it may be considered that different user is to same thing Judge the specific interest tendency shown during opinion.Therefore, the embodiment of the present invention can be by NLP technologies in different comments Appearance analyzed, obtain comment on content similitude either emotion uniformity, if also, targeted customer be directed to first user The similitude or emotion uniformity of the comment content of same article are higher, then the element of correspondence comment behavior in behavior vector Weights are higher, if similitude or emotion uniformity of the targeted customer with first user for the comment content of same article are got over Low, then the weights of the element of correspondence comment behavior are lower in behavior vector.
So, if user A and user B is favorable comment for article a comment, and user C is poor to article a comment Comment, then in the case of other conditions identical, user A and user will be higher than by calculating obtained user A and user B similarity Similarity between C, has further embodied the similarity otherness between different user.
Similarity between S104, each object in the destination object and the object set carries out article and pushed away Recommend.
Seen from the above description, the above method is when calculating similarity, not only allow for destination object and the first object it Between common article, and consider destination object and the first object is directed to the concrete behavior that the article had, add similar The particulate degree judged is spent, the difference between user or article, and different user has been reacted to same article interest level Difference, so as to recommend more to meet the article of demand for user, lift Consumer's Experience.
Above-mentioned steps S104 is described in detail below.Wherein, the embodiment of the present invention recommends article in specially user When, multiple articles can be selected to be recommended from the article set of the other users more similar to targeted customer.
Illustratively, Fig. 2 recommends the method for sequence for a kind of article provided in an embodiment of the present invention, as illustrated, this method Including：
S201, article to be recommended is determined from the article set of other objects similar to the destination object.
For example, user's set of user A concerns includes { B, C, D }, then the embodiment of the present invention can be selected and user A most phases As two users article set in select article to be recommended.If for example, the similarity between user A and user B is more than Similarity between user A and user C, the similarity between user A and user C is more than similar between user A and user D Degree, then can select article to be recommended from user B and user C article set.The embodiment of the present invention can also with Family A similarity, which is more than in the article set of the user of predetermined threshold value, selects article to be recommended, wherein, the threshold value can basis Actual demand is configured, and the present invention is not limited.
The interest level of S202, the calculating destination object to each article to be recommended.
S203, recommendation sequence carried out to each article to be recommended according to the interest level.
Alternatively, the embodiment of the present invention can calculate the destination object to each described to be recommended by equation below The interest level of article：
Wherein, p*(u, i) represents interest levels of the destination object u to article i, and S (u, K) is included and destination object u is emerging The immediate K object of interest, N (i) is the object set for having behavior to article i, wuvIt is the destination object u and described first Object v similarity, riIt is significance levels of the article i in the article to be recommended, nhIt is that the K object had to article h The number of times of behavior, Φ is the article to be recommended, NjIt is the number of times that object set had behavior to article j, Ω is all articles.
Illustratively, object set includes user A, user B, user C and user D, wherein, object set is to all items The number of times for having behavior is 10000000, also, user A article collection is combined into { a, b, d }, user B article collection be combined into a, C }, user C article collection is combined into { b, e }, and user D article collection is combined into { c, d, e }.Also, the method step according to Fig. 1 Suddenly, calculate after the similarity obtained in user A and object set between each user, determine user B in object set, user C And user D is similar to user A.
So, can be from the user B similar to user A, user C and user D when carrying out article recommendation to user A Article to be recommended is selected in article set.Specifically, because user A does not have behavior to article c and article e, and user B and use Family D has behavior to article c, and user C and user D have behavior to article e, hence, it can be determined that article c and article e is to be recommended Article.
Further, if the behavior of user B, user C and user D to article c and article e respectively is as follows：
B：C (is browsed, 1；Comment, 1；Share, 3)
C：E (is browsed, 1；Comment, 1；Share, 5；Thumb up, 1)
D：C (is browsed, 1；Comment, 1；Share, 2；Collection, 1；Thumb up, 1), e (is browsed, 1；Share, 2；Collection, 1)；
All users are as shown in the table to the number of times of the article c and article e behaviors carried out：
And for article c, after the calculating by above-mentioned steps S103, obtain the similarity and use between user A and user B Similarity sum between family A and user D is：
For article e, the similarity sum between similarity and user A and user D between user A and user C is：
It can thus be concluded that, significance levels of the article c in article to be recommended is：
Significance levels of the article e in article to be recommended be：
Then user A is to article c interest level：
User A is to article e interest level：
Therefore, user A is higher than interest levels of the user A to article e to article c interest level, and then can be User's A preferential recommendation articles c.
Using computational methods of the above-mentioned user to article interest level, due to consideration that behavior of the user for article Vector, adds the particulate degree of similarity judgement, has reacted difference of the same user to different article interest levels, so that It can recommend more to meet the article of demand for user priority, lift Consumer's Experience.
The embodiment of the present invention also provides a kind of personalized recommendation device 30, and the one kind provided for embodiment of the method is personalized Recommendation method, as shown in figure 3, the device includes：
First determining unit 301, for determining the first article set of destination object and the second article collection of the first object Identical article among conjunction, first object is any object in object set；
Second determining unit 302, for determining the destination object and first object respectively for the article Behavior vector；
Computing unit 303, for according to the behavior is vectorial, the first article set and the second article collection are total Calculate the similarity between the destination object and first object；
Article recommendation unit 304, between each object in the destination object and the object set Similarity carries out article recommendation.
Above-mentioned personalized recommendation device 30 is not only allowed between destination object and the first object altogether when calculating similarity Same article, and the concrete behavior that destination object and the first object had for the article is considered, add similarity and sentence Disconnected particulate degree, has reacted the difference between user or article, and different user to the difference of same article interest level It is different, so as to recommend more to meet the article of demand for user, lift Consumer's Experience.
Alternatively, second determining unit 302 specifically for：
Similarity between the destination object and first object is calculated according to equation below：
Wherein, N (u) represents the first article set of the destination object u, and N (v) represents the first object v's The second article set, Ωu∩vIt is identical article among the first article set and the second article set,It is The destination object is vectorial for article t behavior,It is behavior vector of first object for article t, w is described Similarity.
The specific description that can refer in above method embodiment for step S103, here is omitted.
Alternatively, a kind of behavior of each element correspondence in the behavior vector, the value of each element is corresponding behavior Weights, wherein, the weights are that the comment content for the article is analyzed by natural language processing NLP technologies Obtained comment content similarities either emotion uniformity, or, the weights are the confidence levels of comment behavior, or, institute It is the numerical value obtained based on term frequency-inverse document frequency TF-IDF statistics to state weights.
Alternatively, the article recommendation unit 304 is used for：
Article to be recommended is determined from the article set of other objects similar to the destination object；
Calculate interest level of the destination object to each article to be recommended；
Recommendation sequence is carried out to each article to be recommended according to the interest level.
Description of the above method embodiment to Fig. 2 specifically is can refer to, here is omitted.
Alternatively, the article recommendation unit 304 is used for：
Interest level of the destination object to each article to be recommended is calculated by equation below：
Wherein, p*(u, i) represents interest levels of the destination object u to article i, and S (u, K) is included and destination object u is emerging The immediate K object of interest, N (i) is the object set for having behavior to article i, wuvIt is the destination object u and described first Object v similarity, riIt is significance levels of the article i in the article to be recommended, nhIt is article h in the article to be recommended In had the number of times of behavior, Φ is the article to be recommended, NjIt is the number of times that article j had behavior in all articles, Ω is All articles.
What deserves to be explained is, the dividing elements that the above is carried out to personalized recommendation device 30, only a kind of logic function is drawn Point, there can be other dividing mode when actually realizing.Also, the physics realization of above-mentioned each functional unit may also have a variety of realities Existing mode.For example, above-mentioned each functional unit can be by software, hardware or both is implemented in combination with the portion as central processing unit Divide or whole.
In addition, affiliated recognize it will be apparent to those skilled in the art ground, and for convenience and simplicity of description, foregoing description Each unit specific work process, may be referred to the corresponding process in preceding method embodiment, here is omitted.
In embodiment provided herein, it should be understood that disclosed apparatus and method, others can be passed through Mode is realized.For example, each functional unit in each embodiment of the invention can be integrated in a processing unit, can also It is that unit is individually physically present.Above-mentioned integrated unit can both be realized in the form of hardware, it would however also be possible to employ hardware Plus the form of SFU software functional unit is realized.
The above-mentioned integrated unit realized in the form of SFU software functional unit, can be stored in an embodied on computer readable and deposit In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions are to cause a computer Equipment (can be personal computer, server, or network equipment etc.) performs the portion of each embodiment methods described of the invention Step by step.And foregoing storage medium includes：(Random Access Memory, arbitrary access is deposited by USB flash disk, mobile hard disk, RAM Reservoir), magnetic disc or CD etc. are various can be with the medium of data storage.
The foregoing is only a specific embodiment of the invention, but protection scope of the present invention is not limited thereto, any Those familiar with the art the invention discloses technical scope in, the change or replacement that can be readily occurred in, all should It is included within the scope of the present invention.Therefore, protection scope of the present invention should be defined by scope of the claims.

Claims (10)

1. a kind of personalized recommendation method, it is characterised in that including：
Determine identical article, described first among the first article set of destination object and the second article set of the first object Object is any object in object set；
Determine the behavior vector of the destination object and first object respectively for the article；
According to the behavior is vectorial, the first article set and the second article set calculate the destination object with it is described Similarity between first object；
Similarity between each object in the destination object and the object set carries out article recommendation.
2. according to the method described in claim 1, it is characterised in that described according to the behavior is vectorial, the first article collection The similarity closed between the second article set calculating destination object and first object includes：
Similarity between the destination object and first object is calculated according to equation below：
w = ( Σ t v , t u ∈ Ω u ∩ v ( a t v · a t u | | a t v | | × | | a t u | | ) ) / | N ( u ) ∪ N ( v ) |
Wherein, N (u) represents the first article set of the destination object u, and N (v) represents that the first object v's is described Second article set, Ωu∩vIt is identical article among the first article set and the second article set,It is institute Behavior vector of the destination object for article t is stated,It is behavior vector of first object for article t, w is the phase Like degree.
3. method according to claim 1 or 2, it is characterised in that each element correspondence in the behavior vector is a kind of Behavior, the value of each element is the weights of corresponding behavior, wherein, the weights are by natural language processing NLP technologies pair Comment content for the article carries out analyzing obtained comment content similarities either emotion uniformity, or, it is described Weights are the confidence levels of comment behavior, or, the weights are obtained based on term frequency-inverse document frequency TF-IDF statistics Numerical value.
4. method according to claim 1 or 2, it is characterised in that described according to the destination object and the object set Similarity between each object in conjunction carries out article and recommends to include：
Article to be recommended is determined from the article set of other objects similar to the destination object；
Calculate interest level of the destination object to each article to be recommended；
Recommendation sequence is carried out to each article to be recommended according to the interest level.
5. method according to claim 4, it is characterised in that the calculating destination object is to each described to be recommended Article interest level, including：
Interest level of the destination object to each article to be recommended is calculated by equation below：
p * ( u , i ) = r i Σ v ∈ S ( u , K ) ∩ N ( i ) w u v
r i = n i Σ h ∈ Φ n h × log Σ j ∈ Ω N j N i
Wherein, p*(u, i) represents interest levels of the destination object u to article i, and S (u, K) is included and destination object u interest most connects K near object, N (i) is the object set for having behavior to article i, wuvIt is the destination object u and the first object v Similarity, riIt is significance levels of the article i in the article to be recommended, nhIt is that the K object had behavior to article h Number of times, Φ is the article to be recommended, NjIt is the number of times that the object set had behavior to article j, Ω is all articles.
6. a kind of personalized recommendation device, it is characterised in that including：
First determining unit, for determining phase among the first article set of destination object and the second article set of the first object Same article, first object is any object in object set；
Second determining unit, for determine the destination object and first object respectively for the article behavior to Amount；
Computing unit, for according to the behavior is vectorial, described in the first article set and the second article set calculate Similarity between destination object and first object；
Article recommendation unit, enters for the similarity between each object in the destination object and the object set Row article is recommended.
7. device according to claim 6, it is characterised in that second determining unit is used for：
Similarity between the destination object and first object is calculated according to equation below：
w = ( Σ t v , t u ∈ Ω u ∩ v ( a t v · a t u | | a t v | | × | | a t u | | ) ) / | N ( u ) ∪ N ( v ) |
Wherein, N (u) represents the first article set of the destination object u, and N (v) represents that the first object v's is described Second article set, Ωu∩vIt is identical article among the first article set and the second article set,It is described Destination object is vectorial for article t behavior,It is behavior vector of first object for article t, w is described similar Degree.
8. the device according to claim 6 or 7, it is characterised in that each element correspondence in the behavior vector is a kind of Behavior, the value of each element is the weights of corresponding behavior, wherein, the weights are by natural language processing NLP technologies pair Comment content for the article carries out analyzing obtained comment content similarities either emotion uniformity, or, it is described Weights are the confidence levels of comment behavior, or, the weights are obtained based on term frequency-inverse document frequency TF-IDF statistics Numerical value.
9. the device according to claim 6 or 7, it is characterised in that the article recommendation unit is used for：
Article to be recommended is determined from the article set of other objects similar to the destination object；
Calculate interest level of the destination object to each article to be recommended；
Recommendation sequence is carried out to each article to be recommended according to the interest level.
10. method according to claim 9, it is characterised in that the article recommendation unit is used for：
Interest level of the destination object to each article to be recommended is calculated by equation below：
p * ( u , i ) = r i Σ v ∈ S ( u , K ) ∩ N ( i ) w u v
r i = n i Σ h ∈ Φ n h × log Σ j ∈ Ω N j N i
Wherein, p*(u, i) represents interest levels of the destination object u to article i, and S (u, K) is included and destination object u interest most connects K near object, N (i) is the object set for having behavior to article i, wuvIt is the destination object u and the first object v Similarity, riIt is significance levels of the article i in the article to be recommended, nhIt is that the K object had behavior to article h Number of times, Φ is the article to be recommended, NjIt is the number of times that the object set had behavior to article j, Ω is all articles.
CN201710108664.3A 2017-02-27 2017-02-27 Personalized recommendation method and device Active CN106952111B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710108664.3A CN106952111B (en) 2017-02-27 2017-02-27 Personalized recommendation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710108664.3A CN106952111B (en) 2017-02-27 2017-02-27 Personalized recommendation method and device

Publications (2)

Publication Number Publication Date
CN106952111A true CN106952111A (en) 2017-07-14
CN106952111B CN106952111B (en) 2021-04-02

Family

ID=59466950

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710108664.3A Active CN106952111B (en) 2017-02-27 2017-02-27 Personalized recommendation method and device

Country Status (1)

Country Link
CN (1) CN106952111B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108737859A (en) * 2018-05-07 2018-11-02 华东师范大学 Video recommendation method based on barrage and device
CN110969222A (en) * 2018-09-29 2020-04-07 西门子股份公司 Information providing method and system
WO2020215743A1 (en) * 2019-04-25 2020-10-29 北京三快在线科技有限公司 Information recommendation

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101916286A (en) * 2010-08-23 2010-12-15 宇龙计算机通信科技(深圳)有限公司 Information recommendation method and system
CN103744966A (en) * 2014-01-07 2014-04-23 Tcl集团股份有限公司 Item recommendation method and device
CN104239338A (en) * 2013-06-19 2014-12-24 阿里巴巴集团控股有限公司 Information recommendation method and information recommendation device
CN104778237A (en) * 2015-04-03 2015-07-15 电子科技大学 Individual recommending method and system based on key users
US20160364783A1 (en) * 2014-06-13 2016-12-15 Truecar, Inc. Systems and methods for vehicle purchase recommendations

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101916286A (en) * 2010-08-23 2010-12-15 宇龙计算机通信科技(深圳)有限公司 Information recommendation method and system
CN104239338A (en) * 2013-06-19 2014-12-24 阿里巴巴集团控股有限公司 Information recommendation method and information recommendation device
CN103744966A (en) * 2014-01-07 2014-04-23 Tcl集团股份有限公司 Item recommendation method and device
US20160364783A1 (en) * 2014-06-13 2016-12-15 Truecar, Inc. Systems and methods for vehicle purchase recommendations
CN104778237A (en) * 2015-04-03 2015-07-15 电子科技大学 Individual recommending method and system based on key users

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
FENG XIE ETC.: "Item Similarity Learning Methods for Collaborative Filtering Recommender Systems", 《2015 IEEE 29TH INTERNATIONAL CONFERENCE ON ADVANCED INFORMATION NETWORKING AND APPLICATIONS》 *
蔡浩 等: "结合用户信任模型的协同过滤推荐方法研究", 《计算机工程与应用》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108737859A (en) * 2018-05-07 2018-11-02 华东师范大学 Video recommendation method based on barrage and device
CN110969222A (en) * 2018-09-29 2020-04-07 西门子股份公司 Information providing method and system
WO2020215743A1 (en) * 2019-04-25 2020-10-29 北京三快在线科技有限公司 Information recommendation

Also Published As

Publication number Publication date
CN106952111B (en) 2021-04-02

Similar Documents

Publication Publication Date Title
CN105701191B (en) Pushed information click rate estimation method and device
US10095771B1 (en) Clustering and recommending items based upon keyword analysis
CN105095267B (en) Recommendation method and device for user participation items
Kumar et al. Social popularity based SVD++ recommender system
EP2304619B1 (en) Correlated information recommendation
CN106651542A (en) Goods recommendation method and apparatus
CN109658206A (en) Information recommendation method and device
CN110335123B (en) Commodity recommendation method, system, computer readable medium and device based on social e-commerce platform
CN110134783B (en) Personalized recommendation method, device, equipment and medium
CN106952111A (en) Personalized recommendation method and device
CN104809637A (en) Commodity recommending method and system realized by computer
CN106776697A (en) Content recommendation method and device
CN104615631A (en) Information recommendation method and device
CN104346428A (en) Information processing apparatus, information processing method, and program
Al Boni et al. Model adaptation for personalized opinion analysis
CN107679236A (en) A kind of hot content pond maintaining method and device
Miao et al. Identifying the most influential topic-sensitive opinion leaders in online review communities
Harahap et al. Laptop selection decision support system according to buyer criteria with the simple additive weighting method
Xu et al. Learning to recommend with user generated content
Ebina et al. Consumer confusion from price competition and excessive product attributes under the curse of dimensionality
Gupta et al. Scalable online product recommendation engine based on implicit feature extraction domain
Mohammadnezhad et al. An effective model for improving the quality of recommender systems in mobile e-tourism
Jain et al. E-commerce product recommendation based on product specification and similarity
CN105512914B (en) Information processing method and electronic equipment
Hu et al. Utilizing users' tipping points in E-commerce Recommender systems

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