CN109978657A - A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms - Google Patents
A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms Download PDFInfo
- Publication number
- CN109978657A CN109978657A CN201910170656.0A CN201910170656A CN109978657A CN 109978657 A CN109978657 A CN 109978657A CN 201910170656 A CN201910170656 A CN 201910170656A CN 109978657 A CN109978657 A CN 109978657A
- Authority
- CN
- China
- Prior art keywords
- article
- user
- similarity
- indicate
- probability
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/06—Asset management; Financial planning or analysis
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- General Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- Marketing (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Human Resources & Organizations (AREA)
- Operations Research (AREA)
- Technology Law (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention proposes a kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms.This method is focused primarily on article network, according to the similarity between the article of the obtained collaborative filtering based on time effect, construct article network, using similarity as the weight of migration, multiple migration can excavate contacting between article and article, go out more similar articles for searching articles, overcome the sparsity of data, promotes rich and coverage rate.The addition of weight is but also recommendation results more personalized.
Description
Technical field
A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms belongs to the field of data mining.
Background technique
Current many data there is certain sparsity, traditional algorithm be difficult to excavate accurate similar users or
Person similar article is predicted and is recommended.Have for the behavioral data of user in data set because finance product quantity is more
There is no co-user between a little finance products, this, which will lead to be difficult to target item, finds article similar enough.But in reality
It, may also phase according to the transitivity of similarity relation, between them if two articles share same similar article in the scene of border
Seemingly.Therefore, by excavating the indirect similarity relation between article, the sparsity that can effectively alleviate direct similarity relation is asked
Topic.
Figure is that a kind of level of abstraction is high, articulate data structure, it is described by the definition to node and side
Incidence relation between entity and entity.User behavior is a natural network, and side and node often have various abundant
Information, figure has transmission capacity, by random walk can excavate spend relationships, can effectively promote coverage, expansion is called together
It returns.
In recent years, the random walk in graph model achieves preferable effect on overcoming data sparsity problem.It can
To regard that a description random walk person accesses the Markov chain of vertex sequence as, when migration person accesses the Ma Er of vertex sequence
Can husband's chain, after migration reaches stable state, the accessed probability of each node is the score of the node.
The step of prior art:
(1) bipartite model is constructed
Model (graph-based model) based on figure is the important content in recommender system.In research based on figure
Before model, it is necessary first to which user behavior data is expressed as to the form of figure.Our user behavior data is by a series of two
Tuple composition, wherein each binary group (u, i) indicates that user u generated behavior to article i.Such data set is held very much
An easy-to-use bipartite graph indicates.
As shown in Figure 1, it is a simple consumer articles bipartite model, wherein circular node represents user, rectangular
Node on behalf article, the side between circular node and square nodes represent behavior of the user to article.Such as user node A in figure
It is connected with article node a, b, d, illustrates that user A generated behavior to article a, b, d.
(2) access probability of article is obtained
In previous step, the behavior representation of user is bipartite model by we.It is in next step exactly on bipartite graph to user
Carry out personalized recommendation.If personalized recommendation algorithm is put into bipartite model, recommend the task of article to user u
The article node correlation on the diagram that measure user vertex v u can be converted into and do not have side to be connected directly with vu, correlation
Weight of the higher article in recommendation list is higher.
The method of correlation between two vertex in measurement bipartite graph is introduced first.Mainly include following three because
Element:
The quantity in path between (1) two vertex.Number of paths between two vertex is more, illustrates the phase of this opposite vertexes
Closing property is just higher.
The length in path between (2) two vertex.If the path length connected between two vertex is all shorter,
The correlation on the two vertex is with regard to relatively high.
The vertex that path between (3) two vertex is passed through.If connecting the path between two vertex without going past out
Bigger vertex is spent, then their correlation also can be relatively high.
It cites a plain example, as shown in Fig. 2 (a), user A does not have side to be connected with article c, e, but user A and object
Product c has the path that 1 length is 3 to be connected, and it is that 3 path is connected that user A, which has 2 length with article e,.So, vertex A and e it
Between correlation be higher than vertex A and c, thus before article e should come article c in the recommendation list of user A, because of top
Have shown in two paths such as Fig. 2 (b) and Fig. 2 (c) between point A and e, is (A, b, C, e) and (A, d, D, e) respectively.Wherein, (A,
B, C, e) path pass through vertex out-degree be (3,2,2,2), and the path (A, d, D, e) pass through vertex out-degree be (3,2,
3,2), the out-degree of certain node refers to the number that the side of other nodes is directed toward by the vertex.Therefore, (A, d, D, e) have passed through one
The bigger vertex D of out-degree, so the contribution of correlation is less than (A, b, C, e) between (A, d, D, e) opposite vertexes A and e.
Based on above-mentioned 3 points, as long as we carry out the calculating of the correlation in figure between vertex.One kind is described below
The most common PersonalRank algorithm based on random walk.
Assuming that personalized recommendation is carried out to user u, from the corresponding node v of user uuStart on consumer articles bipartite graph
Carry out random walk.When migration is to any one node, it is to continue with migration first, in accordance with probability α decision, is also off current trip
It walks and from vuNode starts migration again.If it is determined that continuing migration, then just according to equal from the node that present node is directed toward
The node that even distribution one node of random selection passes through next time as migration.In this way, after random walk many times, each object
The probability that moral integrity point is accessed to can converge to a fixed value.The weight of article is exactly article section in final recommendation list
The access probability of point.Obtain the access probability that following formula calculates article.
Wherein, PR (v) is the probability for accessing vertex v, PR (v ') be access vertex v ' probability, α is to continue with the general of migration
Rate, value are [0,1], and the value of α is determined according to the actual demand of oneself, and 1- α expression rests on vuThe probability of node.General α takes
Value is 0.6 or 0.8, i.e., compared to resting on former vertex, we be more likely to can migration to next vertex increase diversity,
In (v) is directed to the vertex set of vertex v, and v ' is directed to a vertex of vertex v, and out (v ') is vertex v ' direction vertex
Set, | out (v ') | be vertex v ' out-degree.
(3) recommended
According to the access probability of user obtained in the previous step, the recommendation of TopN is carried out from high to low according to access probability.It passes
The Random Walk Algorithm of system, consideration is correlation between user and article, calculates user to article according to the behavior of user
Access probability, and the transition probability between node is set as the same value, in view of different user in recommender system and not
Have the characteristics that different similarities with article.From present node be directed toward node in be according to be uniformly distributed random selection one section
Point as the node passed through migration next time, between vertex all be in this way it is indistinguishable, each vertex is for adjacent vertex
Effect be only to depend on connected side between vertex.When calculating transition probability, due to being jumped to from representative points
Probability between another vertex be it is the same, this, which may will lead to, recommends many unexpected winner articles, reduce conversion ratio.
Recommendation for finance product, the quantity of finance product are much smaller than number of users, and the similitude between article is to us
For it is more valuable, it is intended that recommended by adding the random walk chart-pattern of article similarity weight.Overcome data
Sparsity problem, while promoting click conversion ratio.
Recommendation for finance product, it is contemplated that being more the similitude between article, so it is desirable that passing through
The random walk chart-pattern of the weight of article similarity is added to be recommended.
As shown in figure 3, indicating article figure, the vertex of figure is article, and the value on side indicates the similarity between two articles,
Transition probability i.e. between article.
Summary of the invention
A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms, it is characterised in that the following steps are included:
(1) similarity between article is obtained
The temporal information that user generates behavior to article is obtained, time attenuation function is added to the similarity calculation of article
In formula, formula 3-1 is obtained:
sijIndicate the similarity of article i and article j, u indicates that user, N (i) ∩ N (j) indicate while liking article i and object
The user of product j, | N (i) | it indicates to like the number of users of article i, | N (j) | it indicates to like the number of users of article j;Formula 3-1 is dividing
Introduced in son attenuation term f related with the time (| tui-tuj|);tuiIndicate that user u generates the time of behavior, t to article iujTable
Show user u to article j generate behavior time, f (| tui-tuj|) be meant that, user u generates behavior to article i and article j
Time it is remoter, then f (| tui-tuj|) smaller;f(|tui-tuj|) expression formula such as formula 3-2;
Wherein, e is natural constant, N0=N (0) is N in the initial value at 0 moment, i.e. initial value before time decaying, β
> 0 is known as exponential decay constant;
(2) article adjacency matrix is constructed
According to similarity obtained in (one), 0 all commodity, the adjacent square of building article are greater than for searching articles similarity
Battle array S:
Wherein, sijIndicate the similarity of article i and article j, m is article siiThe quantity of=0 (1≤i≤m), sii=0 (1
≤ i≤m) indicate the similarity for not considering article and its own;
(3) transition probability matrix is updated
By the similarity of article obtained in (one) as the weight on the side of chart-pattern, by the adjacent square of article in 3-3
Battle array S, which is normalized, just obtains transition probability matrix T;Element t in matrixijIt indicates from article i migration to the probability of article j;
Expression formula is
Wherein, sikIndicate the similarity of article i and article k;
(4) article column vector is obtained
R is enabled to indicate article column vector, each element rjIndicate the accessed probability of article j, wherein 1≤i≤m, then at random
The mathematical expression of migration strategy is
rn=c × T × rn-1+(1-c)×r0 (3-5)
Wherein, c is probability of the migration to next node, and 1-c is the probability back to start node i, and the value of c is
[0,1], rnIndicate that the n-th step reaches the probability distribution of each article node, r0Indicate initial probability distribution, its each element rj
Value such as formula 3-6, i.e. the migration since article j, the different degree for initially assigning itself is 1, the different degree of remaining article is 0;
After multiple iteration, article column vector r can converge to a static probability distribution, be
rn=(1-c) × (I-c × T)-1×r0 (3-7)
Wherein, I indicates that unit matrix uses r thus to obtain the similarity between article and articleijIndicate article i and article
The similarity of j;
(5) recommended based on model
Upper section has obtained the similarity between article, in conjunction with the behavioral data of user, calculates user to the inclined of each article
Good value;User u is calculated to the interest of an article j by following formula:
puj=∑I ∈ N (u) ∩ S (i, K)puirij (3-8)
Wherein, pujUser u is indicated to the interest of article j, N (u) indicates the article set that user u likes, and i is user happiness
Some joyous article, S (i, K) is indicated and K article i most like article set, and j is some object in this set
Product, puiIndicate interest of the user u to article i, rijIndicate the similarity of article i and article j;Finally, according to obtained preference
Value, takes TopN to recommend user.
Transition probability between node is set as definite value by traditional Random Walk Algorithm, not in view of different in recommender system
User and different articles have the characteristics that different similarities.Recommendation for finance product, it is contemplated that be more article it
Between similitude, so we are by adding the random walk chart-pattern of the weight of article similarity to recommend.
This patent proposes a kind of proposed algorithm of chart-pattern based on improved random walk.This method is focused primarily on object
On product network, according to the similarity between the article of the obtained collaborative filtering based on time effect, article network is constructed, by phase
Weight like degree as migration, multiple migration, depth excavate contacting between article and article, are that searching articles are more out
Similar article overcomes the sparsity of data, promotes rich and coverage rate.The addition of weight is but also recommendation results have more individual character
Change.
This patent mainly excavates similitude between article, so, the figure that we construct is the article network of Weight, and node is
Commodity, while being similar relationship between article, weight is to joined the similarity of the article of time decaying.
It is because the conventional methods such as random walk are not suitable for article network, article node that we, which need the article figure of Weight,
Very much, the relevance of wherein most node is very weak, that is, unexpected winner article is in the majority, only the figure of small part article building
It is hot spot, if the similarity of many unexpected winner nodes can be made higher, so we are based on side using the method for random walk
Weight is gone migration (weighted walk), and the probability for recommending out compared with hot product is improved.
Detailed description of the invention
Fig. 1 consumer articles bipartite model
The example of proposed algorithm of the Fig. 2 based on figure
The article figure of Fig. 3 Weight
Specific embodiment
(1) similarity between article is obtained
The similarity that the article of behavior occurs within being separated by the very short time for user is higher.By taking finance product is recommended as an example,
The similarity for the finance product that the finance product and user that user buys today were bought yesterday is greater than user in statistical significance
The similarity for the finance product that the finance product of purchase today and user buy the year before.
Our available users generate the temporal information of behavior to article, time attenuation function are added to the phase of article
Like in degree calculation formula, formula 3-1 is obtained:
sijIndicate the similarity of article i and article j, u indicates that user, N (i) ∩ N (j) indicate while liking article i and object
The user of product j, | N (i) | it indicates to like the number of users of article i, | N (j) | it indicates to like the number of users of article j.Formula 3-1 is dividing
Introduced in son attenuation term f related with the time (| tui-tuj|)。tuiIndicate that user u generates the time of behavior, t to article iujTable
Show user u to article j generate behavior time, f (| tui-tuj|) be meant that, user u generates behavior to article i and article j
Time it is remoter, then f (| tui-tuj|) smaller.f(|tui-tuj|) expression formula such as formula 3-2.
Wherein, e is natural constant, N0=N (0) is N in the initial value at 0 moment, i.e., the initial value before time decaying, it
Value according to different in different systems, β > 0 is known as exponential decay constant.Its value is different in different scenes.
If the interests change of user is quickly, bigger β should be just taken, otherwise needs to take smaller β.For example, in finance product
Field, according to the characteristic of finance product, the initial value before time decaying is set as the temperature of finance product, at the beginning of counting finance product
The average reciprocal number of phase is 1000 (as unit of days), and initial value N is arranged0It is 1, the average life period of finance product is
119.65 days, in the end of life of finance product, average reciprocal number only had in 10 (as unit of days), normalization obtain f (|
tui-tuj|) it is 0.01, it brings formula 3-2 into and is fitted to obtain β=0.0385.
(2) article adjacency matrix is constructed
According to similarity obtained in (one), 0 all commodity, the adjacent square of building article are greater than for searching articles similarity
Battle array S:
Wherein, sijIndicate the similarity of article i and article j, m is article siiThe quantity of=0 (1≤i≤m), sii=0 (1
≤ i≤m) indicate the similarity for not considering article and its own.
(3) transition probability matrix is updated
We by the similarity of article obtained in (one) as the weight on the side of chart-pattern, the article in 3-3 is adjacent
It meets matrix S and is normalized and can be obtained by transition probability matrix T.Element t in matrixijIt indicates from article i migration to article j
Probability.Expression formula is
Wherein, sikIndicate the similarity of article i and article k.
(4) article column vector is obtained
R is enabled to indicate article column vector, each element rjIndicate the accessed probability of article j, wherein 1≤i≤m, then at random
The mathematical expression of migration strategy is
rn=c × T × rn-1+(1-c)×r0 (3-5)
Wherein, c is probability of the migration to next node, and 1-c is the probability back to start node i, and the value of c is
The value of [0,1], c is bigger, indicates that we are more likely to migration to next node, will increase the diversity and coverage rate of article,
Value can be carried out according to the actual needs of oneself, such as in the recommendation of finance product, it is therefore desirable to be able to excavate article with more
Association between more articles, so the value of c is 0.8.rnIndicate that the n-th step reaches the probability distribution of each article node, r0It indicates
Initial probability distribution, its each element rjValue such as formula 3-6, i.e. the migration since article j initially assigns the important of itself
Degree is 1, and the different degree of remaining article is 0.
According to above-mentioned strategy, after multiple iteration, article column vector r can converge to static probability point
Cloth is
rn=(1-c) × (I-c × T)-1×r0 (3-7)
Wherein, I indicates unit matrix, and thus we obtain the similarity between article and article, use rijIndicate article i and
The similarity of article j.
(5) recommended based on model
We have obtained the similarity between article to upper section, in conjunction with the behavioral data of user, calculate user to each article
Preference value.User u is calculated to the interest of an article j by following formula:
puj=∑I ∈ N (u) ∩ S (i, K)puirij (3-8)
Wherein, pujUser u is indicated to the interest of article j, (i is user happiness to the article set that N (u) expression user u likes
Some joyous article), S (i, K) indicates that (j is some object in this set with K article i most like article set
Product), puiIndicate interest of the user u to article i, rijIndicate the similarity of article i and article j.The formula is meant that: and use
The family more similar article of interested article in history, more it is possible that obtaining relatively high ranking in the recommendation list of user.
Finally, taking TopN to recommend user according to obtained preference value.
Claims (1)
1. a kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms, it is characterised in that the following steps are included:
(1) similarity between article is obtained
The temporal information that user generates behavior to article is obtained, time attenuation function is added to the calculating formula of similarity of article
In, obtain formula 3-1:
sijIndicate the similarity of article i and article j, u indicates that user, N (i) ∩ N (j) indicate while liking article i's and article j
User, | N (i) | it indicates to like the number of users of article i, | N (j) | it indicates to like the number of users of article j;Formula 3-1 is in the molecule
Introduce attenuation term f related with the time (| tui-tuj|);tuiIndicate that user u generates the time of behavior, t to article iujIt indicates to use
Family u to article j generate behavior time, f (| tui-tuj|) be meant that, user u to article i and article j generate behavior when
Between it is remoter, then f (| tui-tuj|) smaller;f(|tui-tuj|) expression formula such as formula 3-2;
Wherein, e is natural constant, N0=N (0) is N in the initial value at 0 moment, i.e. initial value before time decaying, β > 0 claims
For exponential decay constant;
(2) article adjacency matrix is constructed
According to similarity obtained in (one), 0 all commodity are greater than for searching articles similarity, construct article adjacency matrix S:
Wherein, sijIndicate the similarity of article i and article j, m is article siiThe quantity of=0 (1≤i≤m), sii=0 (1≤i≤
M) similarity for not considering article and its own is indicated;
(3) transition probability matrix is updated
By the similarity of article obtained in (one) as the weight on the side of chart-pattern, by the article adjacency matrix S in 3-3 into
Row normalization just obtains transition probability matrix T;Element t in matrixijIt indicates from article i migration to the probability of article j;Expression formula
For
Wherein, sikIndicate the similarity of article i and article k;
(4) article column vector is obtained
R is enabled to indicate article column vector, each element rjIndicate the accessed probability of article j, wherein 1≤i≤m, then random walk plan
Slightly mathematical expression be
rn=c × T × rn-1+(1-c)×r0 (3-5)
Wherein, c is probability of the migration to next node, and 1-c is the probability back to start node i, and the value of c is [0,1],
rnIndicate that the n-th step reaches the probability distribution of each article node, r0Indicate initial probability distribution, its each element rjValue
Such as formula 3-6, the i.e. migration since article j, the different degree for initially assigning itself is 1, and the different degree of remaining article is 0;
After multiple iteration, article column vector r can converge to a static probability distribution, be
rn=(1-c) × (I-c × T)-1×r0 (3-7)
Wherein, I indicates that unit matrix uses r thus to obtain the similarity between article and articleijIndicate article i's and article j
Similarity;
(5) recommended based on model
Upper section has obtained the similarity between article, in conjunction with the behavioral data of user, calculates user to the preference value of each article;
User u is calculated to the interest of an article j by following formula:
puj=∑I ∈ N (u) ∩ S (i, K)puirij (3-8)
Wherein, pujUser u is indicated to the interest of article j, N (u) indicates the article set that user u likes, and i is that the user likes
Some article, S (i, K) is indicated and K article i most like article set, and j is some article in this set,
puiIndicate interest of the user u to article i, rijIndicate the similarity of article i and article j;Finally, being taken according to obtained preference value
TopN recommends user.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910170656.0A CN109978657A (en) | 2019-03-07 | 2019-03-07 | A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910170656.0A CN109978657A (en) | 2019-03-07 | 2019-03-07 | A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109978657A true CN109978657A (en) | 2019-07-05 |
Family
ID=67078170
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910170656.0A Pending CN109978657A (en) | 2019-03-07 | 2019-03-07 | A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109978657A (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110532528A (en) * | 2019-08-27 | 2019-12-03 | 掌阅科技股份有限公司 | Books similarity calculating method and electronic equipment based on random walk |
CN111104606A (en) * | 2019-12-06 | 2020-05-05 | 成都理工大学 | Weight-based conditional wandering chart recommendation method |
CN111143684A (en) * | 2019-12-30 | 2020-05-12 | 腾讯科技(深圳)有限公司 | Artificial intelligence-based generalized model training method and device |
CN111369306A (en) * | 2020-06-01 | 2020-07-03 | 北京搜狐新媒体信息技术有限公司 | Product recommendation method and device |
CN111488488A (en) * | 2020-04-10 | 2020-08-04 | 杭州趣维科技有限公司 | User interest detection method based on graph mining |
CN112214793A (en) * | 2020-09-30 | 2021-01-12 | 南京邮电大学 | Random walk model recommendation method based on fusion of differential privacy |
CN112364252A (en) * | 2021-01-12 | 2021-02-12 | 北京世纪好未来教育科技有限公司 | Content recommendation method and device, electronic equipment and storage medium |
CN113158053A (en) * | 2021-04-25 | 2021-07-23 | 平安科技(深圳)有限公司 | Service product recommendation method and device, computer equipment and storage medium |
US20220223245A1 (en) * | 2021-01-13 | 2022-07-14 | Boe Technology Group Co., Ltd. | Drug recommendation method, apparatus and system, electronic device and storage medium |
CN115952359A (en) * | 2023-03-10 | 2023-04-11 | 特斯联科技集团有限公司 | Recommendation system recall method and device, electronic equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104391849A (en) * | 2014-06-30 | 2015-03-04 | 浙江大学苏州工业技术研究院 | Collaborative filtering recommendation method for integrating time contextual information |
US20160253325A1 (en) * | 2014-12-30 | 2016-09-01 | Socialtopias, Llc | Method and apparatus for programmatically adjusting the relative importance of content data as behavioral data changes |
CN107590243A (en) * | 2017-09-14 | 2018-01-16 | 中国人民解放军信息工程大学 | The personalized service recommendation method to be sorted based on random walk and diversity figure |
-
2019
- 2019-03-07 CN CN201910170656.0A patent/CN109978657A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104391849A (en) * | 2014-06-30 | 2015-03-04 | 浙江大学苏州工业技术研究院 | Collaborative filtering recommendation method for integrating time contextual information |
US20160253325A1 (en) * | 2014-12-30 | 2016-09-01 | Socialtopias, Llc | Method and apparatus for programmatically adjusting the relative importance of content data as behavioral data changes |
CN107590243A (en) * | 2017-09-14 | 2018-01-16 | 中国人民解放军信息工程大学 | The personalized service recommendation method to be sorted based on random walk and diversity figure |
Non-Patent Citations (2)
Title |
---|
QWUREY: "[推荐算法]ItemCF,基于物品的协同过滤算法", 《HTTPS://BLOG.CSDN.NET/YERUBY/ARTICLE/DETAILS/44154009》 * |
赵海燕等: "基于相似性随时间衰减的矩阵分解算法", 《小型微型计算机系统》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110532528A (en) * | 2019-08-27 | 2019-12-03 | 掌阅科技股份有限公司 | Books similarity calculating method and electronic equipment based on random walk |
CN111104606B (en) * | 2019-12-06 | 2022-10-21 | 成都理工大学 | Weight-based conditional wandering chart recommendation method |
CN111104606A (en) * | 2019-12-06 | 2020-05-05 | 成都理工大学 | Weight-based conditional wandering chart recommendation method |
CN111143684A (en) * | 2019-12-30 | 2020-05-12 | 腾讯科技(深圳)有限公司 | Artificial intelligence-based generalized model training method and device |
CN111143684B (en) * | 2019-12-30 | 2023-03-21 | 腾讯科技(深圳)有限公司 | Artificial intelligence-based generalized model training method and device |
CN111488488A (en) * | 2020-04-10 | 2020-08-04 | 杭州趣维科技有限公司 | User interest detection method based on graph mining |
CN111369306A (en) * | 2020-06-01 | 2020-07-03 | 北京搜狐新媒体信息技术有限公司 | Product recommendation method and device |
CN112214793A (en) * | 2020-09-30 | 2021-01-12 | 南京邮电大学 | Random walk model recommendation method based on fusion of differential privacy |
CN112364252A (en) * | 2021-01-12 | 2021-02-12 | 北京世纪好未来教育科技有限公司 | Content recommendation method and device, electronic equipment and storage medium |
US20220223245A1 (en) * | 2021-01-13 | 2022-07-14 | Boe Technology Group Co., Ltd. | Drug recommendation method, apparatus and system, electronic device and storage medium |
CN113158053B (en) * | 2021-04-25 | 2022-09-30 | 平安科技(深圳)有限公司 | Service product recommendation method and device, computer equipment and storage medium |
CN113158053A (en) * | 2021-04-25 | 2021-07-23 | 平安科技(深圳)有限公司 | Service product recommendation method and device, computer equipment and storage medium |
CN115952359A (en) * | 2023-03-10 | 2023-04-11 | 特斯联科技集团有限公司 | Recommendation system recall method and device, electronic equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109978657A (en) | A kind of improvement random walk chart-pattern proposed algorithm towards many intelligence platforms | |
Yu et al. | A cross-domain collaborative filtering algorithm with expanding user and item features via the latent factor space of auxiliary domains | |
Teo et al. | Adaptive, personalized diversity for visual discovery | |
CN103678672B (en) | Method for recommending information | |
CN107590243B (en) | The personalized service recommendation method to be sorted based on random walk and diversity figure | |
US20230153857A1 (en) | Recommendation model training method, recommendation method, apparatus, and computer-readable medium | |
CN109299370A (en) | Multipair grade personalized recommendation method | |
CN111428147A (en) | Social recommendation method of heterogeneous graph volume network combining social and interest information | |
CN109993583B (en) | Information pushing method and device, storage medium and electronic device | |
Zhang et al. | Deep sequential model for anchor recommendation on live streaming platforms | |
CN112989169B (en) | Target object identification method, information recommendation method, device, equipment and medium | |
CN102214169A (en) | Methods and devices for providing keyword information and target information | |
CN110851651B (en) | Personalized video recommendation method and system | |
CN109190030A (en) | Merge the implicit feedback recommended method of node2vec and deep neural network | |
Xia et al. | Predicting user replying behavior on a large online dating site | |
Liu et al. | User preference modeling based on meta paths and diversity regularization in heterogeneous information networks | |
CN109684561B (en) | Interest point recommendation method based on deep semantic analysis of user sign-in behavior change | |
CN106156113A (en) | A kind of method of video recommendations, system and electronic equipment | |
US8849830B1 (en) | Delivering search results | |
GB2576938A (en) | System and method for improved content discovery | |
CN111079005B (en) | Recommendation method based on item time popularity | |
CN110969535A (en) | Method, device, system and medium for matching between users | |
CN108763515A (en) | A kind of time-sensitive personalized recommendation method decomposed based on probability matrix | |
Ezra et al. | Prophet inequality with competing agents | |
Kochubey et al. | Modified algorithm of collaborative filtering for forming user recommendations |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190705 |
|
RJ01 | Rejection of invention patent application after publication |