CN107391670A - A kind of mixing recommendation method for merging collaborative filtering and user property filtering - Google Patents
A kind of mixing recommendation method for merging collaborative filtering and user property filtering Download PDFInfo
- Publication number
- CN107391670A CN107391670A CN201710599815.XA CN201710599815A CN107391670A CN 107391670 A CN107391670 A CN 107391670A CN 201710599815 A CN201710599815 A CN 201710599815A CN 107391670 A CN107391670 A CN 107391670A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- user
- mover
- msup
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of mixing for merging collaborative filtering and user property filtering to recommend method, and it is to recommend according to user's rating matrix to user to solve traditional Collaborative Filtering Recommendation Algorithm, it is sparse there is factor data and the problem of cause recommendation information inaccuracy.The present invention is improved using the computational methods of time temperature to Pearson correlation coefficient algorithm, then establishes user property similarity model, neighbor user is filtered, active user is recommended by the trusted neighbor finally given.Test result indicates that the more traditional system filter algorithm of mixing proposed algorithm proposed by the present invention has more preferable effect.
Description
Technical field
The present invention relates to a kind of mixing for merging collaborative filtering and user property filtering to recommend method, belongs to information technology neck
Domain.
Background technology
At present, the annual training of Yunnan Power System reaches 60,000 person-times, has training scale big, training contents broad covered area, specially
The features such as industry is more, strongly professional, how its project (such as equipment interested pushed to user according to the behavioral data of each user
Knowledge point) become training in a problem.Wherein, the main part of commending system is personalized recommendation algorithm, and research pushes away
It is exactly to study personalized recommendation algorithm in fact to recommend system, because the performance of commending system is dependent on the property of personalized recommendation algorithm
Energy.The proposed algorithm used in current all kinds of commending systems has a many kinds, wherein collaborative filtering be current application most
Extensively, the personalized recommendation algorithm of most study.Collaborative filtering is found out the similar users of active user and utilized similar first
User is recommended, and the self attributes without considering project, the opinion for depending on nearest-neighbors user is recommended, and is inclined to
In personalized recommendation.
Collaborative filtering biggest advantage is that do not have special requirement to recommended, can handle non-structured complexity
Object, the major defect of collaborative filtering is Sparse sex chromosome mosaicism, i.e., in user's rating matrix data than in the case of sparse,
The Similarity Measure of user is inaccurate, and the effect for causing to recommend thus can not be met into system requirements.Researcher is in order to enter
One step improves the recommendation quality of collaborative filtering, for its shortcoming, employs a series of processing method, such as《Based on unusual
The Collaborative Filtering Recommendation Algorithm research that value is decomposed》Elaborate how to utilize the mode of singular value decomposition to reduce the dimension of rating matrix
Number, useful information is obtained, but this method has lacked partial data, and in Sparse, it is traditional that it predicts that error can be more than
Collaborative filtering.《Collaborative Filtering Recommendation Algorithm based on filling and similitude trust-factor》Elaborate that one kind carries out scoring square
The data mining algorithm of battle array filling, the algorithm has been partially improved collaborative filtering, but the algorithm excessively relies on any active ues,
When scoring sparse, it is unfavorable for carrying out personalized recommendation to user.And traditional Collaborative Filtering Recommendation Algorithm is commented according to user
Sub-matrix is recommended to user, it is sparse there is factor data and the problem of cause recommendation information inaccuracy.
Recommending field, collaborative filtering is practical proposed algorithm, and the algorithm is divided to two kinds：One kind is based on use
The collaborative filtering (UserCF) at family；Also one kind is project-based collaborative filtering (ItemCF).Based on user's
The general principle of collaborative filtering recommending is：Preference according to all users to article, find and active user's taste and preference phase
As neighbor user group, be then based on the history preference information of neighbours group, recommended for active user.Project-based collaboration
The general principle of filtered recommendation is：According to all users to article preference, the similarity between article and article, Ran Hougen are found
According to the history preference information of user, similar article is recommended into user.
The present invention fusion collaborative filtering and user property filter mixing proposed algorithm advantage be mainly reflected in by with
Family attribute filtering is applied in the searching of collaborative filtering trusted neighbor, so can effectively be alleviated because rating matrix data are dilute
The defects of searching similar users brought are inaccurate is dredged, meanwhile, it is excellent by being carried out to Pearson correlation coefficient in collaborative filtering
Change, also increase the confidence level for finding similar users, recommendation accuracy rate can be effectively improved.
The content of the invention
To achieve the above object, the present invention provides following technical scheme：One kind fusion collaborative filtering and user property filtering
Mixing recommend method, it comprises the following steps：
(1) Pearson correlation coefficient algorithm is improved after obtaining improvement using the computational methods of time heat degree function
Calculating formula of similarity；
(2) similarity of active user and other users is calculated according to the calculating formula of similarity after the improvement of step (1),
And find out the N number of preliminary neighbor user of final similarity highest；
(3) user property similarity model is established, neighbor user is further filtered, obtains final trusted neighbor collection
Close M；
(4) active user is recommended by the trusted neighbor finally given.
Further, preferably, in the step (1), time heat degree function calculation formula is as follows：
Wherein, if Dui represents that user u accesses project i time and user u was accessed between any one of system object time earliest
Every the time interval has the corresponding time to record in database, and time heat degree function WT (u, i) is one related to Dui
Functional value, it uses the non-decreasing function on Dui, i.e., for Dui ＞ Duj, there is WT (u, i) >=WT (u, j), the time temperature
Function is a line shape function, and wherein Lu represents that user u uses the time span of commending system, i.e. the user accesses system earliest
Any one of any one object time and nearest access system object time interval, a ∈ (0,1), referred to as weight growth indices, change
A value can adjust the speed that weight changes over time, and a is bigger, and weight growth rate is faster, and a big I has influence on algorithm
Can, the value that dynamic adjusts a optimizes recommendation effect.
Further, preferably, in the step (1), Pearson correlation coefficient algorithm is specially：
User a and user b Pearson similarities represent as follows：
Wherein, give user and collect rating matrix R, ra, the p expression user a of U, Item Sets P and user to project to project
P scoring,Represent the average value that user u scores project P.
Further, preferably, in the step (1), time temperature is added to calculating formula of similarity, optimized
Calculating formula of similarity after the improvement obtained afterwards is as follows：
Wherein, introduce after time temperature, when calculating a and b similarity, the recent interest of user will reflect more
Fully, the similarity of active user and remaining user can be calculated using the formula after Optimal improvements, and according to Top-N principles
Select the N positions neighbours of active user.
Further, preferably, filtering out the low neighbours of similarity using user property, and the attribute of user is carried out
Feature extraction forms eigenmatrix, and the similarity between user is calculated using eigenmatrix.
Further, preferably, after the eigenmatrix of user is established, first, N is compared by the calculation formula after improvement
N positions neighbours are resequenced by position neighbours and the similarity of active user according to similarity is descending, select preceding M positions most
Whole trusted neighbor, wherein M ＜ N；
Afterwards, active user is calculated by M positions trusted neighbor to score to the unknown purpose, wherein, user a is pre- to project p's
Test and assess and divide ra, p calculation formula is as follows：
The final recommended project collection of the high item design active user of prediction scoring is selected according to Top-N principles.
Compared with prior art, the beneficial effects of the invention are as follows：
The present invention calculates the similarity between user, it is contemplated that user based on the collaborative filtering based on user
Interest can change over time, the present invention propose time temperature concept simultaneously Similarity Measure is optimized, can be effectively
Alleviate the defects of searching similar users brought due to rating matrix Sparse are forbidden, meanwhile, by collaborative filtering
Middle Pearson correlation coefficient optimizes, and also increases the confidence level for finding similar users, can effectively improve recommendation accuracy rate.
Brief description of the drawings
Fig. 1 is the MAE values based on collaborative filtering；
MAE values of the Fig. 2 based on mixing proposed algorithm of the present invention；
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under the premise of creative work is not made
Embodiment, belong to the scope of protection of the invention.
The present invention provides a kind of technical scheme：A kind of mixing recommendation method for merging collaborative filtering and user property filtering,
It is based on the collaborative filtering based on user, calculates the similarity between user.Can be with view of the interest of user
Time change, the present invention propose the concept of time temperature and Similarity Measure are optimized.
1st, user's rating matrix of the present invention is introduced first：
In systems, user can carry out interest scores to project, and score value scope is 1-5, and scoring is higher, represents user
It is bigger to the interest-degree of project.If I1, I2, I3 ..., IM be system project, U1, U2, U3 ..., UN be system user,
Then scoring of the user to project is inserted in corresponding matrix unit, you can obtain user-project rating matrix, such as following table：
1 user of table-project rating matrix
| I1 | I2 | I3 | … | IM | |
| U1 | 5 | - | 1 | … | - |
| U2 | - | - | 2 | … | 4 |
| U3 | 3 | 4 | - | … | - |
| … | … | … | … | … | … |
| UN | 5 | - | - | … | - |
2nd, the time temperature of the lower present invention is introduced
Traditional algorithm does not consider that different user access item purpose has when finding the nearest-neighbors of active user
The body time to predicting the influence of non-scoring item, have ignored user interest can time to time change this rule, in order to find pair
The more valuable similar users of recommendation results, it is contemplated that the project that user accesses in the recent period more can than the project for a long time accessed before
Time temperature is incorporated into calculating formula of similarity by this reason of the interest of reaction user, the present invention, and phase is found to reach increase
Like the purpose of User reliability.
Time temperature refers to user's access item object time freshness, and access time is from the more near then freshness of current time more
Height, time temperature is higher, and vice versa.If Dui represents that user u accesses project i time and user u accesses system and appointed earliest
The time interval (having the corresponding time to record in database) of one project, defines time heat degree function WT (u, i), it is one
The functional value related to Dui.In order to protrude the importance for the project that user u was accessed in the recent period, function is designed to close by the present invention
In Dui non-decreasing function, i.e., for Dui ＞ Duj, there is WT (u, i) >=WT (u, j), time heat degree function calculation formula is as follows：
Above formula is a line shape function, and wherein Lu represents that user u uses the time span of commending system, i.e. the user is earliest
Any one of any one of access system object time and nearest access system object time interval, a ∈ (0,1), referred to as weight increase
Index.The speed that weight changes over time can be adjusted by changing a value.A is bigger, and weight growth rate is faster, and a size can be with
Algorithm performance is had influence on, can dynamically adjust a value to optimize recommendation effect.
3rd, the optimization of Pearson correlation coefficient
The method of conventional calculating user's similarity is Pearson correlation coefficient, gives user and collects U, Item Sets P and use
To the rating matrix R (such as table 1) of project, ra, p represent scorings of the user a to project p at family,Represent that user u scores project P
Average value, then user a and user b similarity represent it is as follows：
Traditional algorithm find active user nearest-neighbors when have ignored user interest can time to time change this
Rule, in order to find the similar users more valuable to recommendation results, the present invention enters to Pearson correlation coefficient (formula (2))
Row improves, and time temperature is added into calculating formula of similarity, the calculating formula of similarity after optimization is as follows：
As can be seen that introducing after time temperature from formula (3), when calculating a and b similarity, user is recent
Interest will reflect more abundant.The similarity of active user and remaining user can be calculated using the formula (3) after optimization, and
And the N positions neighbours of active user are selected according to Top-N principles.
The similar users set being calculated with the improved user's calculating formula of similarity (formula (3)) of the present invention, not
It is can have good recommendation effect to all users, because may exist in set widely different with targeted customer's interest
Similar users, as caused by such similar users recommend accuracy rate be than relatively low.Why this phenomenon can be had, it is main
If because rating matrix than what is caused by sparse reason, next done seeks to filter out this kind of similarity ratio again
Relatively low user.Main method is to establish its user property model for each user, by compare user property similarity come
Filtering.
4th, user property filters out the low neighbours of similarity
The low neighbours of similarity are filtered out using user property, it is necessary to which the attribute progress feature extraction to user forms spy
Matrix is levied, the similarity between user is calculated using eigenmatrix.
1) eigenmatrix of user is established
One user can have a more attribute, such as work post, educational background, sex, and present invention extraction is wherein compared with can react user
7 attribute of feature build user characteristics matrix, and this 7 attribute is respectively：Work post, educational background, the length of service, relevant department, sex,
Post, grade of skill.
Eigenmatrix is as shown in table 2 below：
The user characteristics matrix of table 2
| User | Work post | Grade of skill | Post | Relevant department | … |
| User 1 | Work post 1 | Grade 1 | Post 1 | Department 1 | … |
| User 2 | Work post 2 | Grade 1 | Post 2 | Department 2 | … |
| User 3 | Work post 3 | Grade 1 | Post 1 | Department 1 | … |
| … | … | … | … | … | … |
2) similarity between user is calculated
User characteristics attribute includes work post, educational background, the length of service, relevant department, sex, post, grade of skill, then user u
Characteristic attribute can be represented with vectorial UAttru=(au1, au2, au3, au4, au5, au6, au7).Wherein, divide from u1 to u7
Table represents work post, educational background, the length of service, relevant department, sex, post, grade of skill.For numerical attribute, such as length of service, according to reality
If experience present invention provide that the two length of service differ by more than 3 years old, then it is assumed that the two is different；For categorical attribute, such as work post,
Go through, relevant department, sex, post, grade of skill, use original value, if user u is identical with user v ith attribute, we
USimUAttr (u, v, i)=1 is made, otherwise USimUAttr (u, v, i)=0.User u and v similarity can use following public affairs
Formula calculates
USimAttr (u, v)=∑i∈UAttrωi·UsimUAttr(u, v, i) (4)
In formula：ωiFor the weight of ith attribute, the weighted value of all properties adds up to 1.
5th, the description of recommendation step
The mixing of fusion collaborative filtering and user property filtering recommends method specific implementation flow to have following steps：
1. the specific time of project accessible by user, time temperature is calculated according to formula (1).
2. for user to be recommended, according to the calculating formula of similarity (formula (3)) improved calculate active user with
The similarity of other users, the preliminary similar neighborhood collection being made up of N positions user is then drawn using Top-N method.
3. establish the eigenmatrix of user according to the methods that 3.2 sections are introduced, and by formula (4) compare N positions neighbours with
The similarity of active user, N positions neighbours are resequenced according to similarity is descending, final credible in M positions before selecting
Neighbours (M ＜ N).
4. by M positions trusted neighbor active user is calculated to score to the unknown purpose, wherein, pre- test and appraisal of the user a to project p
Point ra, p calculation formula are as follows：
The final recommended project collection of the high item design active user of prediction scoring is selected according to Top-N principles.
6th, experiment and interpretation of result
1) experimental data and measurement
Sparse degree refers to the relative percentage of unit and total unit not comprising data, and its calculation formula is as follows：
In formula：A expressions have included the unit number of data, and P represents total unit number
The data set that the present invention uses picks up from《Based on digitally coded mobile learning management platform》Caused data set, should
Data set includes 8600 interest scores of 897 users to 122 projects, and the value of scoring is 1 to 5, can according to formula (6)
Sparse degree is calculated as 0.9214.80% is randomly selected to this data set and is used as training set, residue 20% is used as test set.
The prediction of all units in test set scoring is calculated using the data in training set and the algorithm belonging to the present invention, is then contrasted
Actual scoring in test set can be to algorithm recommendation quality analyze.
Experiment is used as evaluation index using mean absolute error (MAE), and MAE measures prediction of the user to project in test set
Scoring and the error actually to score, MAE is smaller, illustrates to recommend quality higher.Assuming that the user of prediction scores, set expression is
{ P1, P2 ... PN }, corresponding actual user's scoring collection are combined into { q1, q2 ..., qN }, then specific MAE calculation formula are
2) analysis of experimental results
In order to verify the validity of mixing proposed algorithm of the present invention, respectively to traditional collaborative filtering (UserCF) and
Mixing proposed algorithm (Hybrid Recommendation Method, HRM) of the present invention has carried out contrast experiment, the result of experiment
As shown in Figure 1-2.Abscissa is K values (number of users) in figure, and ordinate is evaluation index MAE values.
(1) can be drawn from Fig. 1-2, the MAE values based on collaborative filtering will be more than the present invention in whole k values section and mix
The MAE values of proposed algorithm are closed, MAE is smaller, represents to recommend quality higher, mixing proposed algorithm of the present invention thus can be explained
Integrally recommending to be better than traditional collaborative filtering on precision.
(2) can be drawn from two figures, after k ＞ 60, with the increase of k values, the MAE values of two kinds of algorithms all increased,
But the rate of rise of the MAE values based on collaborative filtering is apparently higher than the MAE values based on mixing proposed algorithm of the present invention
Rate of rise, MAE value rates of rise are lower, then it represents that recommend stability better, recommendation of the present invention thus can be explained and calculate
Method is better than traditional collaborative filtering in stability.
The present invention is from the angle for finding trusted neighbor, it is proposed that a kind of fusion collaborative filtering and user property filtering
Mixing proposed algorithm, analyzed by parameters relationship and recommend method contrast experiment to show that mixing proposed algorithm of the invention is
Effective algorithm, its recommendation effect are better than traditional collaborative filtering.
Although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of changes, modification can be carried out to these embodiments, replace without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (6)
1. a kind of mixing recommendation method for merging collaborative filtering and user property filtering, it comprises the following steps：
(1) phase after being improved is improved to Pearson correlation coefficient algorithm using the computational methods of time heat degree function
Like degree calculation formula；
(2) similarity of active user and other users is calculated according to the calculating formula of similarity after the improvement of step (1), and is looked for
Go out the N number of preliminary neighbor user of final similarity highest；
(3) user property similarity model is established, neighbor user is further filtered, obtains final trusted neighbor set M；
(4) active user is recommended by the trusted neighbor finally given.
2. a kind of mixing recommendation method for merging collaborative filtering and user property filtering according to claim 1, its feature
It is：In the step (1), time heat degree function calculation formula is as follows：
<mrow>
<mi>W</mi>
<mi>T</mi>
<mrow>
<mo>(</mo>
<mi>u</mi>
<mo>,</mo>
<mi>i</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mi>a</mi>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msup>
<mi>a</mi>
<mfrac>
<mrow>
<mi>D</mi>
<mi>u</mi>
<mi>i</mi>
</mrow>
<mrow>
<mi>L</mi>
<mi>u</mi>
</mrow>
</mfrac>
</msup>
</mrow>
Wherein, if Dui represents that user u accesses project i time and user u accesses any one of system object time interval earliest,
The time interval has the corresponding time to record in database, and time heat degree function WT (u, i) is a function related to Dui
Value, it uses the non-decreasing function on Dui, i.e., for Dui ＞ Duj, there is WT (u, i) >=WT (u, j), the time heat degree function
It is a line shape function, wherein Lu represents that user u uses the time span of commending system, i.e. it is any to access system earliest by the user
Item any one of object time and nearest access system object time interval, a ∈ (0,1), referred to as weight growth indices, changes a's
Value can adjust the speed that weight changes over time, and a is bigger, and weight growth rate is faster, and a big I has influence on algorithm performance,
Dynamic adjustment a value optimizes recommendation effect.
3. a kind of mixing recommendation method for merging collaborative filtering and user property filtering according to claim 2, its feature
It is：In the step (1), Pearson correlation coefficient algorithm is specially：
User a and user b Pearson similarities represent as follows：
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<mi>P</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>a</mi>
<mo>.</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>a</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>b</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>b</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msqrt>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<mi>P</mi>
</mrow>
</msub>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>a</mi>
<mo>.</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>a</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<msqrt>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<mi>P</mi>
</mrow>
</msub>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>b</mi>
<mo>.</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>b</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mrow>
</mfrac>
</mrow>
Wherein, give user and collect rating matrix R, ra, the p expression user a of U, Item Sets P and user to project to project p's
Scoring,Represent the average value that user u scores project P.
4. a kind of mixing recommendation method for merging collaborative filtering and user property filtering according to claim 3, its feature
It is：In the step (1), time temperature is added to calculating formula of similarity, the phase after the improvement obtained after optimization
It is as follows like degree calculation formula：
<mrow>
<msup>
<mi>sim</mi>
<mo>*</mo>
</msup>
<mrow>
<mo>(</mo>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<mi>P</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mrow>
<mi>W</mi>
<mi>T</mi>
<mrow>
<mo>(</mo>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>a</mi>
<mo>.</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>a</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<mrow>
<mi>W</mi>
<mi>T</mi>
<mrow>
<mo>(</mo>
<mrow>
<mi>b</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>b</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>b</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msqrt>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<mi>P</mi>
</mrow>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>W</mi>
<mi>T</mi>
<mrow>
<mo>(</mo>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>a</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<msqrt>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>p</mi>
<mo>&Element;</mo>
<mi>P</mi>
</mrow>
</msub>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>W</mi>
<mi>T</mi>
<mrow>
<mo>(</mo>
<mrow>
<mi>b</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
<mo>)</mo>
</mrow>
<mo>&times;</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>b</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>b</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
</mrow>
</mfrac>
</mrow>
Wherein, introduce after time temperature, when calculating a and b similarity, the recent interest of user, which will reflect, more fills
Point, the similarity of active user and remaining user can be calculated using the formula after Optimal improvements, and select according to Top-N principles
Go out the N positions neighbours of active user.
5. a kind of mixing recommendation method for merging collaborative filtering and user property filtering according to claim 4, its feature
It is：The low neighbours of similarity are filtered out using user property, and feature extraction formation feature square is carried out to the attribute of user
Battle array, the similarity between user is calculated using eigenmatrix.
6. a kind of mixing recommendation method for merging collaborative filtering and user property filtering according to claim 5, its feature
It is：After the eigenmatrix of user is established, first, N positions neighbours and active user are compared by the calculation formula after improvement
Similarity, N positions neighbours are resequenced according to similarity is descending, the final trusted neighbor in M positions, wherein M before selecting
＜ N；
Afterwards, active user is calculated by M positions trusted neighbor to score to the unknown purpose, wherein, pre- test and appraisal of the user a to project p
Point ra, p calculation formula are as follows：
<mrow>
<msub>
<mi>r</mi>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>=</mo>
<mover>
<msub>
<mi>r</mi>
<mi>a</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>+</mo>
<mfrac>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>b</mi>
<mo>&Element;</mo>
<mi>M</mi>
</mrow>
</msub>
<msup>
<mi>sim</mi>
<mo>*</mo>
</msup>
<mo>(</mo>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
</mrow>
<mo>)</mo>
<mo>&times;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>r</mi>
<mrow>
<mi>b</mi>
<mo>.</mo>
<mi>p</mi>
</mrow>
</msub>
<mo>-</mo>
<mover>
<msub>
<mi>r</mi>
<mi>b</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<msub>
<mi>&Sigma;</mi>
<mrow>
<mi>b</mi>
<mo>&Element;</mo>
<mi>M</mi>
</mrow>
</msub>
<msup>
<mi>sim</mi>
<mo>*</mo>
</msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>a</mi>
<mo>,</mo>
<mi>b</mi>
</mrow>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
The final recommended project collection of the high item design active user of prediction scoring is selected according to Top-N principles.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710599815.XA CN107391670A (en) | 2017-07-21 | 2017-07-21 | A kind of mixing recommendation method for merging collaborative filtering and user property filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710599815.XA CN107391670A (en) | 2017-07-21 | 2017-07-21 | A kind of mixing recommendation method for merging collaborative filtering and user property filtering |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107391670A true CN107391670A (en) | 2017-11-24 |
Family
ID=60336683
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710599815.XA Pending CN107391670A (en) | 2017-07-21 | 2017-07-21 | A kind of mixing recommendation method for merging collaborative filtering and user property filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107391670A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460145A (en) * | 2018-03-15 | 2018-08-28 | 北京邮电大学 | A kind of collaborative filtering recommending method based on mixing Interest Similarity |
| CN108681947A (en) * | 2018-05-21 | 2018-10-19 | 辽宁师范大学 | The collaborative filtering recommending method of association in time degree and coverage based on article |
| CN108829819A (en) * | 2018-06-12 | 2018-11-16 | 上海智臻智能网络科技股份有限公司 | Personalized text recommended method and system, server, readable storage medium storing program for executing |
| CN108830460A (en) * | 2018-05-23 | 2018-11-16 | 重庆邮电大学 | A method of recommender system Deta sparseness is alleviated based on substep dynamic filling |
| CN108876508A (en) * | 2018-05-03 | 2018-11-23 | 上海海事大学 | A kind of electric business collaborative filtering recommending method |
| CN109241203A (en) * | 2018-09-27 | 2019-01-18 | 天津理工大学 | A kind of user preference and distance weighted clustering method of time of fusion factor |
| CN111486345A (en) * | 2020-03-10 | 2020-08-04 | 安徽科杰粮保仓储设备有限公司 | Grain depot underground pipe network liquid leakage on-line monitoring and early warning method and device |
| CN111523054A (en) * | 2020-04-26 | 2020-08-11 | 山东师范大学 | Project recommendation method and system based on active account and similar accounts |
| CN113392320A (en) * | 2021-05-14 | 2021-09-14 | 国家卫星气象中心(国家空间天气监测预警中心) | Knowledge service-based satellite remote sensing service providing method |
| CN116049573A (en) * | 2023-03-28 | 2023-05-02 | 南京邮电大学 | User hierarchical recommendation method for improving collaborative filtering |
-
2017
- 2017-07-21 CN CN201710599815.XA patent/CN107391670A/en active Pending
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460145A (en) * | 2018-03-15 | 2018-08-28 | 北京邮电大学 | A kind of collaborative filtering recommending method based on mixing Interest Similarity |
| CN108460145B (en) * | 2018-03-15 | 2020-07-03 | 北京邮电大学 | Collaborative filtering recommendation method based on mixed interest similarity |
| CN108876508A (en) * | 2018-05-03 | 2018-11-23 | 上海海事大学 | A kind of electric business collaborative filtering recommending method |
| CN108681947B (en) * | 2018-05-21 | 2021-09-24 | 辽宁师范大学 | Collaborative filtering recommendation method based on time relevance and coverage of articles |
| CN108681947A (en) * | 2018-05-21 | 2018-10-19 | 辽宁师范大学 | The collaborative filtering recommending method of association in time degree and coverage based on article |
| CN108830460A (en) * | 2018-05-23 | 2018-11-16 | 重庆邮电大学 | A method of recommender system Deta sparseness is alleviated based on substep dynamic filling |
| CN108830460B (en) * | 2018-05-23 | 2021-12-17 | 重庆邮电大学 | Method for relieving data sparsity of recommendation system based on step-by-step dynamic filling |
| CN108829819A (en) * | 2018-06-12 | 2018-11-16 | 上海智臻智能网络科技股份有限公司 | Personalized text recommended method and system, server, readable storage medium storing program for executing |
| CN108829819B (en) * | 2018-06-12 | 2021-05-07 | 上海智臻智能网络科技股份有限公司 | Personalized text recommendation method and system, server and readable storage medium |
| CN109241203A (en) * | 2018-09-27 | 2019-01-18 | 天津理工大学 | A kind of user preference and distance weighted clustering method of time of fusion factor |
| CN109241203B (en) * | 2018-09-27 | 2021-08-31 | 天津理工大学 | Clustering method for user preference and distance weighting by fusing time factors |
| CN111486345A (en) * | 2020-03-10 | 2020-08-04 | 安徽科杰粮保仓储设备有限公司 | Grain depot underground pipe network liquid leakage on-line monitoring and early warning method and device |
| CN111523054A (en) * | 2020-04-26 | 2020-08-11 | 山东师范大学 | Project recommendation method and system based on active account and similar accounts |
| CN113392320A (en) * | 2021-05-14 | 2021-09-14 | 国家卫星气象中心(国家空间天气监测预警中心) | Knowledge service-based satellite remote sensing service providing method |
| CN116049573A (en) * | 2023-03-28 | 2023-05-02 | 南京邮电大学 | User hierarchical recommendation method for improving collaborative filtering |
| CN116049573B (en) * | 2023-03-28 | 2023-09-01 | 南京邮电大学 | User hierarchical recommendation method for improving collaborative filtering |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107391670A (en) | A kind of mixing recommendation method for merging collaborative filtering and user property filtering | |
| CN110532436B (en) | Cross-social network user identity recognition method based on community structure | |
| CN104063481B (en) | A kind of film personalized recommendation method based on the real-time interest vector of user | |
| CN109299380B (en) | Exercise personalized recommendation method based on multi-dimensional features in online education platform | |
| CN107562947A (en) | A kind of Mobile Space-time perceives the lower dynamic method for establishing model of recommendation service immediately | |
| CN102214241B (en) | Method for detecting burst topic in user generation text stream based on graph clustering | |
| CN105373597A (en) | Collaborative filtering recommendation method for user based on k-medoids project clustering and local interest fusion | |
| CN104317900A (en) | Multiattribute collaborative filtering recommendation method oriented to social network | |
| CN107679239B (en) | Personalized community recommendation method based on user behaviors | |
| CN103678618A (en) | Web service recommendation method based on socializing network platform | |
| CN105354260B (en) | The Mobile solution of a kind of mosaic society's network and item characteristic recommends method | |
| CN112598438A (en) | Outdoor advertisement recommendation system and method based on large-scale user portrait | |
| CN104021153A (en) | Personalized recommendation method for campus books | |
| CN104462383A (en) | Movie recommendation method based on feedback of users' various behaviors | |
| CN103823888A (en) | Node-closeness-based social network site friend recommendation method | |
| CN106817251A (en) | A kind of link prediction method and device based on node similarity | |
| CN105678590A (en) | topN recommendation method for social network based on cloud model | |
| CN105653640A (en) | Collaborative filtering recommendation method based on trust mechanism | |
| CN105095476A (en) | Collaborative filtering recommendation method based on Jaccard equilibrium distance | |
| CN109919459B (en) | Method for measuring influence among social network objects | |
| CN115270007B (en) | POI recommendation method and system based on mixed graph neural network | |
| CN111324807A (en) | Collaborative filtering recommendation method based on trust degree | |
| CN116244513A (en) | Random group POI recommendation method, system, equipment and storage medium | |
| CN106649731A (en) | Node similarity searching method based on large-scale attribute network | |
| CN104731887B (en) | A kind of user method for measuring similarity in collaborative filtering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171124 |
|
| WD01 | Invention patent application deemed withdrawn after publication |