CN114140170A - Intelligent recommendation system based on user preference - Google Patents

Intelligent recommendation system based on user preference Download PDF

Info

Publication number
CN114140170A
CN114140170A CN202111498122.4A CN202111498122A CN114140170A CN 114140170 A CN114140170 A CN 114140170A CN 202111498122 A CN202111498122 A CN 202111498122A CN 114140170 A CN114140170 A CN 114140170A
Authority
CN
China
Prior art keywords
query
preference
user
score
attribute
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
Application number
CN202111498122.4A
Other languages
Chinese (zh)
Inventor
王习特
韩玉雪
周虹宇
白梅
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.)
Dalian Maritime University
CERNET Corp
Original Assignee
Dalian Maritime University
CERNET 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 Dalian Maritime University, CERNET Corp filed Critical Dalian Maritime University
Priority to CN202111498122.4A priority Critical patent/CN114140170A/en
Publication of CN114140170A publication Critical patent/CN114140170A/en
Pending legal-status Critical Current

Links

Images

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
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0255—Targeted advertisements based on user history
    • G06Q30/0256—User search
    • 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
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0254—Targeted advertisements based on statistics
    • 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
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0259—Targeted advertisements based on store location
    • 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
    • G06Q30/0241—Advertisements
    • G06Q30/0251—Targeted advertisements
    • G06Q30/0269—Targeted advertisements based on user profile or attribute
    • 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
    • G06Q30/0241—Advertisements
    • G06Q30/0277—Online advertisement

Landscapes

  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Engineering & Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Probability & Statistics with Applications (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses an intelligent recommendation system based on user preference, which comprises: the system comprises a query grouping module, a query processing module and a user preference acquisition module; the query grouping module is used for grouping the user query requests; the user preference acquisition module is used for initializing preference information and dynamically adjusting the preference information; and the query processing module is used for processing the grouped query requests, performing sorting and screening according to the preference scores calculated by the user preference acquisition module, and returning the sorted and screened query results to the user. The invention can dynamically adjust the preference of the user, can better determine the user requirement, saves the user selection time, and can also dynamically adjust the preference setting along with the change of the user preference, thereby reducing the user setting time and keeping the recommendation accuracy. The invention can sort and screen the inquired results, so that the inquired results returned to the user are closer to the user preference, and the user selection time is reduced.

Description

Intelligent recommendation system based on user preference
Technical Field
The invention relates to the technical field of intelligent recommendation, in particular to an intelligent recommendation system based on user preference.
Background
At present, with the rise of network technologies, high-speed internet has been widely deployed and applied, and users can conveniently perform electronic shopping, data retrieval and the like by using the network. However, as the information is explosively increased, users are more and more dependent on effective information retrieval and filtering technology to deal with information overload, so that the research on intelligent result recommendation is hot, and more intelligent and customized recommendation information is provided for the users.
However, in the existing online recommendation system, personal preferences of each user are generally not considered, but a large number of unordered recommendation results are returned to the user, and information lacks effective sorting and filtering, so that the user needs to spend a large amount of time for browsing, but cannot quickly locate information actually needed.
Disclosure of Invention
The invention provides an intelligent recommendation system based on user preference, which aims to overcome the problems.
The invention comprises the following steps: the method comprises the following steps: the system comprises a query grouping module, a query processing module and a user preference acquisition module;
the query grouping module is used for grouping the user query requests;
the user preference acquisition module is used for acquiring user preferences and dynamically adjusting the user preferences to provide recommendation results according with the user preferences for the user;
dynamically adjusting preference information, namely dynamically adjusting the preference information of the user by combining the scores and the historical preferences according to the browsing time and the final selection of the user on the recommendation result; grading the preference information, and grading the shop finally according to the grading of the preference information;
and the query processing module is used for processing the grouped query requests, performing sorting and screening according to the preference scores calculated by the user preference acquisition module to obtain recommendation results, and returning the recommendation results to the user.
Further, the user preference obtaining module scores each item of preference information when dynamically adjusting the preference,
scoring the preference information is based on the following processing strategy:
according to the numerical characteristics and the actual application scene of each preference attribute, the preference attribute types are divided into: a segmented type, a continuous type and a 01 type,
step a, respectively calculating the score of each preference attribute according to the preference attribute type:
A) when the preference attribute is continuous and is in the value interval of the preference attribute selected by the user, judging that the preference attribute is segmented; the score calculation formula of the segment-type preference is as follows:
Figure BDA0003401676510000021
wherein, H.score [ i ] is preference score, [ min, max ] is value range laid on the preference attribute by all shops, and [ min _ pre, max _ pre ] is value section approved by user; q is the actual value of the ith attribute of the H shop; score [ i ] is the preference score for the ith attribute;
B) continuous type, that is, attribute values corresponding to the preference are continuous; the score calculation formula of the continuous preference is as follows:
Figure BDA0003401676510000022
wherein HrTaking the value of the shop in the preference attribute, and taking Fm as the maximum value of the preference attribute;
C) type 01, when the attribute value corresponding to the preference is Boolean type, the attribute value is 1 point if the requirement is met, and the attribute value is 0 point if the requirement is not met, the preference is judged to be type 01; the score calculation formula for type 01 preferences is:
Figure BDA0003401676510000023
step b, calculating the final score of the shop:
Socre＝∑i∈[1,n]w[i]×H.score[i] (4)
wherein Socre is the final score of the shop, i represents any preference attribute of the shop, H.score [ i ] is the score of the ith preference attribute, and w [ i ] is the weight coefficient of the ith preference score
Figure BDA0003401676510000024
Where the argument x [ i ] is an integer and the initial value is set to 0, the initial value of the weight coefficient w [ i ] is therefore 1.
Further, the specific processing strategy for the dynamic adjustment of the preference is as follows:
to adjust the weighting factor w [ i ] of the i-th preference]Presetting a shop browsing time threshold in a fixed time period, and counting the browsing times exceeding the time threshold as ClThe times meeting the requirement of the ith preference of the user are
Figure BDA0003401676510000025
The number of times of the user's order is recorded as CdThe times meeting the requirement of the ith preference of the user are
Figure BDA0003401676510000026
The formula of the weight adjustment decision factor J is:
Figure BDA0003401676510000027
wherein alpha is a proportion coefficient set according to experience and used for adjusting the proportion of each item in the formula;
for the weight coefficient w [ i ]]Setting an increase threshold θ and a decrease threshold
Figure BDA0003401676510000031
If the factor J is greater than the increase threshold theta, indicating that the preference setting meets the actual requirements of the user, then x [ i ] in the formula]Self-increasing 1, w [ i ]]Is correspondingly increased; if the factor J is below the lowering threshold
Figure BDA0003401676510000032
X [ i ] in the formula if the preference setting does not meet the actual requirement of the user]Self-decreasing 1, w [ i ]]Correspondingly decreases; the increase threshold θ and the decrease threshold
Figure BDA0003401676510000033
Are all set according to experience;
for the preference set by the user, if the factor J is lower than the reduction threshold value continuously in a plurality of statistical periods
Figure BDA0003401676510000034
If the preference type is 01 type, suggesting the user to delete the preference, and if the preference type is segmented or continuous type, suggesting the user to modify the preference; for the preference which is not set by the user, if the judgment factor J is larger than the increase threshold theta in a plurality of continuous statistical periods, recommending the user to add the preference which is not set; the statistical period duration is set empirically.
Further, keyword query, namely query of store information close to the given keywords; inquiring the geographic position, namely inquiring shop information in a certain distance range of a given position, wherein the certain distance range is set according to experience;
the query grouping module employs the following processing strategy:
step 1, dividing the query request into two types: a keyword query request and a geographic position query request;
step 2, processing the query request according to the type of the query request:
(A) keyword-oriented query grouping:
step A1, presetting a similarity threshold, calculating the similarity between two keyword requests, if the similarity is greater than or equal to the preset threshold, judging that the two keyword requests are similar, and dividing the similar requests into the same group; the similarity calculation formula between any two keyword requests is as follows:
Figure BDA0003401676510000035
wherein q is1A keyword request indicating that the number of keywords is smaller than the number of keywords in the other keyword request among the two keyword requests; k represents q1Arbitrary key of (1), q1Key represents q1Key word of (1) | q1Key | represents q1Number of keywords in (1), N (k) represents q2Whether there is a synonym for k in the key;
the formula for N (k) is as follows:
Figure BDA0003401676510000036
step A2, scoring the query packet by using a set scoring function f (Q), wherein f (Q) is calculated as follows:
Figure BDA0003401676510000041
where | Qg | represents the number of packets for all requests, QiRepresenting an arbitrary packet request, Sim (Q)i) Representing a packet QiThe calculation formula is as follows:
Figure BDA0003401676510000042
wherein, | QiI represents a packet QiThe number of requests;
a3, if the score of the query group is larger than the threshold value, the group is qualified, and the qualified query group is sent to the query processing module for processing; if the score of the query grouping is smaller than the threshold value, the grouping is unqualified, and the grouping is deleted; the threshold value is set according to experience;
(B) geographic location-oriented query grouping:
step B1, presetting a similarity value threshold, calculating the similarity between any two geographical position query requests, if the similarity is less than or equal to the preset value, judging that the two geographical position query requests are similar, and dividing the similar geographical position query requests into the same group, wherein the calculation formula of the similarity between any two geographical position query requests is as follows:
Figure BDA0003401676510000043
wherein q is1Pos represents q1Position information of (a), q2Pos represents q2The location information of (a); if Sim (q)1,q2) If the geographic position query requests are 1, judging that the two geographic position query requests are similar, and dividing the similar geographic position query requests into the same group; if Sim (q)1,q2) If the two geographic position query requests are not similar to each other, judging that the two geographic position query requests are not similar to each other; the set distance is set according to experience;
step B2, scoring the packet by using a scoring function f (Q);
step B3, if the score of the query group is larger than the threshold value, the group is qualified, and the qualified query group is sent to the query processing module for processing; if the score of the query grouping is smaller than the threshold value, the grouping is unqualified, and the grouping is deleted; the threshold is set empirically.
Further, the query processing module is based on the following processing strategies:
step 1, retrieving the groups sent by the query grouping module, acquiring shops meeting the query keywords, and adding the acquired shops into a query result set:
step 1A, keyword-oriented query is based on the following processing strategies:
step a1, selecting a main query for a plurality of keyword queries in the group, wherein the main query is the query with the highest similarity score with other queries in the group, and the query similarity score calculation formula is as follows:
Figure BDA0003401676510000051
wherein Q isiRepresenting the ith group of keyword queries, qiIs QiThe ith keyword query of the group, Sim (q)iQ) is qiSimilarity to other keyword queries in the group;
step a2, processing the main query: if the keywords contained in the main query completely cover the keywords contained in other queries in the group, returning the result of the main query to other queries in the group; otherwise, merging the obtained main query and other queries which are not contained into a new query, and repeating the step a1 until a main query meeting the convergence condition is obtained, wherein the convergence condition is as follows: the keywords contained in the main query completely cover the keywords contained in other queries in the group;
step a3, retrieving keywords in the main query to obtain shops meeting the query keywords, and adding the obtained shops into a query result set;
step 1B, querying facing to the geographic position, based on the following processing strategies:
b1, merging the query ranges of the plurality of query conditions in the group, scanning the merged query ranges, and extracting all shop results in the ranges;
b2, judging the extracted results of each shop, judging whether the shop meets the query requirement, if so, adding the results into a query result set, wherein the query requirement is set according to experience;
step b3, sending the stores with the concentrated query results to a preference acquisition module, and calculating the store scores;
b4, sorting the query result set based on the user preference: and sequencing the query results according to the shop scores calculated by the preference acquisition module to obtain a top-k result set R, selecting the preferred skyline recommendation from the top-k result set R as a recommendation result, and sending the recommendation result to the user, wherein the top-k result set R is the query result selected according to requirements, and the requirements are set according to experience.
Further, the selection of the preferred skyline is based on the following processing strategy:
for each query result in the top-k result set R, performing skyline filtering based on each preference attribute dimension of the query result, wherein the skyline filtering comprises the following steps: for any two stores H in R1、H2If the shop H1Store H is not less than or equal to each preference attribute2At least one preference attribute is greater than store H2Then call H1Dominating H2(ii) a The dominated stores in the top-k result set R are deleted, and the remaining stores, as preferred skyline points, are returned to the user.
Further, when the user uses the method for the first time, the query keywords input by the user are used as basic preference information, and the background preference information is supplemented according to the group characteristics of the user.
The invention can dynamically adjust the preference of the user, can better determine the user requirement, saves the user selection time, and can also dynamically adjust the preference setting along with the change of the user preference, thereby reducing the user setting time and keeping the recommendation accuracy. The invention can also sort the inquired results and screen the results, so that the inquired results returned to the user are closer to the user preference, and the user selection time is reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a block diagram of the system of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
As shown in fig. 1, the present invention includes: the system comprises a query grouping module, a query processing module and a user preference acquisition module;
the inquiry grouping module is used for grouping user inquiry requests;
the query processing module is used for processing a query request of a user;
the user preference acquisition module is used for acquiring user preferences and dynamically adjusting the user preferences;
preferably, the user preference obtaining module includes: initializing preference information and dynamically adjusting preference;
initializing preference information, namely analyzing the preference information according to a query keyword input by a user when the preference information is used for the first time by the user to serve as basic preference information; and according to the group characteristics of the user, the background preference information is supplemented.
And dynamically adjusting the preference information, namely scoring each item of preference information according to the browsing time and the final selection of the user on the recommendation result, and dynamically adjusting the preference information of the user by combining the scoring and the historical preference.
Preferably, the preference score is based on the following processing strategy:
the preference types are classified into a segmented type, a continuous type and a 01 type according to the numerical characteristics of each preference attribute and the actual application scene.
Specifically, the determination is made based on the attribute of the preference, such as "wifi" preference, and the type 01 is the case only with or without the preference. The "price" preference is usually a certain price interval, such as 300- "400". The preference can be expressed by a segment interval, and is segmented. The scoring preference is generally continuous numerical, and larger is better, continuous.
A) When the preference attribute is continuous and is in the value interval of the preference attribute selected by the user, the preference attribute is judged to be segmented, and the calculation formula of the shop price score recommended by the segmented preference score is as follows:
Figure BDA0003401676510000071
wherein [ min _ pre, max _ pre ] is a price interval preferred by the user; q is the price of the recommended store; [ min, max ] is the store price interval in the local area; score [ i ] is the score of the ith preference;
B) the calculation formula of the shop price score recommended by the continuous preference score is as follows:
Figure BDA0003401676510000072
wherein HrA score for the store; fm is full mark;
C) and type 01, when the attribute value corresponding to the preference is Boolean, if the attribute value meets the requirement, the score is 1, and if the attribute value does not meet the requirement, the score is 0, the preference is judged to be type 01, and the calculation formula of the shop price score recommended by the type 01 preference score is as follows:
Figure BDA0003401676510000073
preferably, for any attribute of the store H, the corresponding preference type is judged firstly, and then the preference score of the attribute is calculated according to the corresponding formula of the type, wherein the formulas are detailed in (1), (2) and (3).
Each item is scored by a weight coefficient, which initially defaults to 1.
The final Score for each result is the sum of the preference scores for the respective attributes, accumulated by weight, and is given by the formula:
Socre＝∑i∈[1,n]w[i]×H.score[i] (4)
where i represents any of the preferences, h.score [ i ] is the score of the ith preference, and w [ i ] is the weight of the ith preference score.
After using the system for a period of time, dynamic adjustment is needed, and the calculation formula of the dynamically adjusted w [ i ] is shown in formula (5)
Preferably, the dynamic adjustment of the preference is based on the following processing strategy:
the weight adjustment w [ i ] borrows the idea of sigmoid function, and controls the ascending and descending upper and lower boundaries through an independent variable x, and the formula is as follows:
Figure BDA0003401676510000081
where the argument x [ i ] is an integer and the initial value is set to 0, the initial value of the weight coefficient w [ i ] is therefore 1. Each time the weight is adjusted up or down, x [ i ] is increased or decreased by 1.
To adjust the weighting factor w [ i ] of the i-th preference]Presetting a shop browsing time threshold in a fixed time period, and counting the browsing times exceeding the time threshold as ClWherein the number of times meeting the i-th preference requirement of the user is recorded as
Figure BDA0003401676510000082
The number of times of the user's order is recorded as CdWherein the number of times meeting the i-th preference requirement of the user is recorded as
Figure BDA0003401676510000083
A weight adjustment decision factor J of
Figure BDA0003401676510000084
Wherein alpha is a specific gravity coefficient set by the system and is used for adjusting the specific gravity of each item in the formula.
Specifically, the representative user is satisfied with the attributes of the shop with respect to the shop that is finally purchased and the shop that browses for more than a certain time, and the representative user reflects the user preference to some extent. Alpha can be set according to needs and used for matching the specific gravity relationship of two shops.
Preferably, for the user-set preference, if the factor J is below the decrease threshold continuously for a plurality of statistical periods
Figure BDA0003401676510000085
If the preference type is 01 type, the user is suggested to delete the preference, and if the preference type is segmented or continuous type, the user is suggested to modify the preference. And for the preference attribute which is not set by the user, if the judgment factor J is larger than the increase threshold value theta in a plurality of continuous statistical periods, recommending the user to add the preference.
The following case is used for explanation:
in a period of using the system, the user is counted to order 9 shops together, and the number of browsing time exceeding 3 minutes is 12. Wifi is found in 7 of the 9 bought stores in the ordering process, and wifi is found in 12 stores with browsing time exceeding 3 minutes and 9 browsing stores. Formula (6) was used at a ratio of 5:1
Figure BDA0003401676510000086
The judgment factor of the wifi weight adjustment is 77.19 percent which is larger than the threshold value of the weight adjustment by 50 percent, and the independent variable x [ i ] in the weight formula (5) with the wifi is used]Automatic 1 up-regulation, increasing from 0 to 1, corresponding
Figure BDA0003401676510000087
Increasing from 1 to 1.46.
Preferably, a threshold value is preset for the preference not selected by the user, and if the occupation ratio is greater than the threshold value in the continuous m periods, the user is recommended to add the preference.
Specifically, m has a value of 3. If the preference list initially has no preference of "being close to the subway", if the stores close to the subway exceed the set threshold value in 3 consecutive periods, the user is considered to be interested in the preference, and the preference is recommended to the user and added to the preference list.
And presetting a threshold value for each type of preference selected by the user, if the proportion of the preference is lower than the set threshold value in continuous m periods, recommending the user to delete the preference if the preference is 01 type, and recommending the user to modify corresponding preference information if the preference is segmented or continuous type, wherein the threshold value is recommended to be in a range of [ 0.1-0.3 ].
Preferably, the grouping module is queried. The system provides two kinds of query interfaces to the user: keyword query, namely query of store information close to a given keyword; geographic location query, i.e., query for store information near a given location. The system receives a large number of inquiry requests in a short time, and for different types of inquiry, the inquiry needs to be grouped in advance, and the following processing strategies are adopted.
Keyword-oriented query request grouping:
presetting a similarity threshold, calculating the similarity between two keyword requests, comparing the calculated similarity with a preset value by the system, if the similarity is larger than the preset value, considering that the two keyword requests are sufficiently similar, dividing the sufficiently similar requests into the same group by the system, and calculating the similarity between any two keyword requests according to the following formula:
Figure BDA0003401676510000091
wherein q is1Indicating a low number of keys in two key requests, k indicating q1Arbitrary key of (1), q1Key represents q1Key word of (1) | q1Key | represents q1Number of keywords in (1), N (k) represents q2Whether there is a synonym for k in the key;
the formula for N (k) is as follows:
Figure BDA0003401676510000092
specifically, assume that there are now two multi-keyword queries, one of which contains: there is wifi, clean, provide breakfast, another keyword includes: neat, wifi, convenient traffic, 24 hours provide the hot water.
Put few key words to q1In (1), q1: has wifi, is clean and provides breakfast
q2Neat, wifi, convenient traffic and 24-hour hot water supply.
Wherein q is1The first keyword comprises wifi and q2The second keyword in the list has wifi, which is a similar word, so that N (with wifi) is 1; q. q.s1Second keyword is clean and q2The first keyword is a synonym, so that N (clean) is 1; q. q.s1The third keyword provides breakfast and q2The keyword in (1) has no similar meaning word, so that N (providing breakfast) is 0; therefore, the method comprises the following steps:
Figure BDA0003401676510000093
and (f) a set scoring function f (Q) is adopted to measure the quality of the packet and complete the query packet, wherein the f (Q) calculation formula is as follows:
Figure BDA0003401676510000101
where | Qg | represents the number of packets for all requests, QiRepresenting an arbitrary packet request, Sim (Q)i) Representing a packet QiThe calculation formula is as follows:
Figure BDA0003401676510000102
wherein, | QiI represents a packet QiNumber of requests in
Grouping facing to the geographic position query request:
presetting a similarity value, calculating the similarity between any two geographic position query requests, comparing the calculated similarity with a preset value by the system, if the similarity is equal to the preset value, determining that the two geographic position query requests are sufficiently similar, dividing the sufficiently similar geographic position query requests into the same group by the system, and calculating the similarity between any two space requests according to a formula:
Figure BDA0003401676510000103
wherein q is1Pos represents q1Position information of (a), q2Pos represents q2The location information of (1).
Specifically, the similarity value is 1, defined according to the similarity calculation formula (11), and if the distance between two locations is less than the threshold value, which is 1, then two space requests are considered to be sufficiently similar to be placed in the same group.
The scoring functions f (Q) set by the formulas (9) and (10) are adopted to measure the quality of the grouping, and the grouping of the query is completed;
preferably, the query processing module includes: concurrent computation and query result set ordering based on user preferences;
the concurrent computation is that for a plurality of query requests in the group, the system processes the query requests according to different query types, and the query types are divided into multi-keyword query processing and space query processing;
concurrent computation of multi-keyword query is based on the following concurrent processing strategies:
1) for a plurality of keyword queries in a group, selecting a main query, namely selecting the main query with a highest similarity score between a certain query in the group and other queries, wherein the similarity score calculation formula is as follows:
Figure BDA0003401676510000104
wherein Q isiRepresenting the ith group of keyword queries, qiIs QiThe ith keyword query of the group, Sim (q)iQ) is qiSimilarity to other queries in the group, the calculation formulae being detailed in (7)
2) And after the main query is obtained, processing the main query, and in the process of processing the main query, if the main query result completely covers other queries in the group, returning the result of the main query to the other queries in the group.
3) Repeating the steps 1) and 2) for the queries which cannot be completely covered in the group until all the queries are calculated;
concurrent computation of the geographic location query is based on the following concurrent processing strategies:
1) merging the query ranges of a plurality of query conditions in the group to form a minimum outsourcing rectangle query range, and performing range search by using the outsourcing rectangle;
2) calculating each query in the group in sequence for each scanned result, judging whether the result meets the query requirement, and if so, adding the result into a query result set;
and (3) sequencing the query result sets based on the preference of the user, namely distributing the query result sets of the group to each query according to the parameters of the queries in the group, sequencing the query results obtained by each query according to the preference scores to obtain a top-k result set R, selecting the preferred skyline from the result set R and recommending the skyline to the user, wherein the top-k is the top 70% of the selected query results.
Preferably, the top-k result set R is based on the following processing strategy:
and respectively matching the group query result set with the queries in the group according to the parameters of the queries in the group, calculating scores of the query results obtained by the queries according to the preference, sorting in a descending order, and returning the first k results as the query result sets R.
Skyline is preferred, based on the following processing strategy:
for each query result in the top-k result set R, performing skyline filtering based on each preference attribute dimension of the query result, wherein the skyline filtering comprises the following steps: for any two stores H in R1、H2If H is1Is no more than H in all dimensions2Poor and at least one dimension better than H2Then call H1Dominating H2. Then all of the dominated stores will be filtered out of the result set R and will remainThe remaining stores, i.e., the preferred skyline points, are returned to the user as the final recommendation.
Specifically, the attribute dimension of preference is each attribute of preference, and mainly includes price of a store, score of a store, window, free air conditioning, proximity to ground iron, breakfast provision, smoking capability, and the like.
The specific comparison is that different judgment criteria exist according to the type of preference. The lower the price, the better the segment type attribute, the higher the store score, the better the continuity, and the better the type 01.
The following examples are used to explain how this is particularly true:
suppose the query result includes store H1，H2：
Windowed, free air conditioned, breakfast served, etc.:
H1the price of the store was 150, the score was 4.9,
windowless, free air conditioning, breakfast serving, etc.:
H2the price of the store was 200, the score was 4.5,
H1price ratio of (H)2Low score ratio H2High and windowed, other preference attribute dimensions and H2In the same way, then H1Dominating H2Is prepared from H2And filtering out.
Example 2
In this embodiment, the user needs to actively input initial preference information when using the system for the first time, and the initial preference information selected when registering an account includes "wifi" and represents a store where the user tends to have wifi. "have wifi" is type 01 preference, and the corresponding weighting factor defaults to 1. Given a browsing time threshold of 3 minutes and a weight up-regulation threshold of 50%, in a period of time using the system, the user is counted to subscribe 9 shops altogether, and the number of times that the browsing time exceeds 3 minutes is 12. Wifi is found in 7 of the 9 bought stores in the ordering process, and wifi is found in 12 stores with browsing time exceeding 3 minutes and 9 browsing stores. When. varies ═ 5, the formula (6) is used
Figure BDA0003401676510000121
Figure BDA0003401676510000122
The judgment factor of the wifi weight adjustment is 77.19 percent which is larger than the threshold value of the weight adjustment by 50 percent, and the independent variable x [ i ] in the weight coefficient (formula 5) with the preference of wifi is added]Automatic 1 up-regulation, increasing from 0 to 1, corresponding
Figure BDA0003401676510000123
Increasing from 1 to 1.46.
In this embodiment, three users initiate the geographic location query (a, 3), (B, 3), (C, 3) at the same time, which means: the query locations are A, B, C, respectively, centered at A, B, C, with a query radius of 3 kilometers. When the query engine receives a plurality of query requests, the query grouping module performs query grouping, and after the similarity between any two queries in the three queries is calculated by the system by using a formula (11), the queries are considered to be similar and are grouped into the same group. The query processing module performs concurrent computation on queries in the group, and the system completes computation results of all queries in the group simultaneously through one-time data reading. First, the center in the group is calculated as (D, 5), and store queries are performed with D as the center and 5 km as the radius, thereby obtaining A, B, C stores that can cover all queries. And in the query grouping module, for each queried shop, sequentially judging whether the shop is in the respective query range according to the query parameters in the groups (A, 3), (B, 3) and (C, 3), and further determining the query result of each query in the group. And (4) scoring and sequencing respective query results according to preference information of the user, and filtering the scored query results according to the characteristics of top-k query and skyline query to obtain an optimal query result which is a result set recommended to the user.
Has the advantages that:
the invention can dynamically adjust the preference of the user, can better determine the user requirement, saves the user selection time, and can also dynamically adjust the preference setting along with the change of the user preference, thereby reducing the user setting time and keeping the recommendation accuracy. The invention can also sort the inquired results and screen the results, so that the inquired results returned to the user are closer to the user preference, and the user selection time is reduced.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (7)

1. An intelligent recommendation system based on user preferences, comprising: the system comprises a query grouping module, a query processing module and a user preference acquisition module;
the query grouping module is used for grouping the user query requests;
the user preference acquisition module is used for acquiring user preferences and dynamically adjusting the user preferences to provide recommendation results according with the user preferences for the user;
dynamically adjusting preference information, namely dynamically adjusting the preference information of the user by combining the scores and the historical preferences according to the browsing time and the final selection of the user on the recommendation result; grading the preference information, and grading the shop finally according to the grading of the preference information;
and the query processing module is used for processing the grouped query requests, performing sorting and screening according to the preference scores calculated by the user preference acquisition module to obtain recommendation results, and returning the recommendation results to the user.
2. The intelligent recommendation system based on user preferences according to claim 1, wherein the user preference obtaining module scores each item of preference information while dynamically adjusting the preferences,
scoring the preference information is based on the following processing strategy:
according to the numerical characteristics and the actual application scene of each preference attribute, the preference attribute types are divided into: a segmented type, a continuous type and a 01 type,
step a, respectively calculating the score of each preference attribute according to the preference attribute type:
A) when the preference attribute is continuous and is in the value interval of the preference attribute selected by the user, judging that the preference attribute is segmented; the score calculation formula of the segment-type preference is as follows:
Figure FDA0003401676500000011
wherein, H.score [ i ] is preference score, [ min, max ] is value range laid on the preference attribute by all shops, and [ min _ pre, max _ pre ] is value section approved by user; q is the actual value of the ith attribute of the H shop; score [ i ] is the preference score for the ith attribute;
B) continuous type, that is, attribute values corresponding to the preference are continuous; the score calculation formula of the continuous preference is as follows:
Figure FDA0003401676500000012
wherein HrValue of the store at this preference attribute, FmThe maximum value of the preference attribute is taken;
C) type 01, when the attribute value corresponding to the preference is Boolean type, the attribute value is 1 point if the requirement is met, and the attribute value is 0 point if the requirement is not met, the preference is judged to be type 01; the score calculation formula for type 01 preferences is:
Figure FDA0003401676500000021
step b, calculating the final score of the shop:
Socre＝∑i∈[1，n]w[i]×H.score[i] (4)
wherein Socre is the final score of the shop, i represents any preference attribute of the shop, H.score [ i ] is the score of the ith preference attribute, and w [ i ] is the weight coefficient of the ith preference score
Figure FDA0003401676500000022
Where the argument x [ i ] is an integer and the initial value is set to 0, the initial value of the weight coefficient w [ i ] is therefore 1.
3. The intelligent recommendation system based on user preferences according to claim 2, wherein the specific processing strategy for the dynamic adjustment of preferences is as follows:
to adjust the weighting factor w [ i ] of the i-th preference]Presetting a shop browsing time threshold in a fixed time period, and counting the browsing times exceeding the time threshold as ClThe times meeting the requirement of the ith preference of the user are
Figure FDA0003401676500000023
The number of times of the user's order is recorded as CdThe times meeting the requirement of the ith preference of the user are
Figure FDA0003401676500000024
The formula of the weight adjustment decision factor J is:
Figure FDA0003401676500000025
wherein alpha is a proportion coefficient set according to experience and used for adjusting the proportion of each item in the formula;
for the weight coefficient w [ i ]]Setting an increase threshold θ and a decrease threshold
Figure FDA0003401676500000026
If the factor J is greater than the increase threshold θ, the preference is indicatedSetting x [ i ] in the formula to meet the actual requirement of the user]Self-increasing 1, w [ i ]]Is correspondingly increased; if the factor J is below the lowering threshold
Figure FDA0003401676500000027
X [ i ] in the formula if the preference setting does not meet the actual requirement of the user]Self-decreasing 1, w [ i ]]Correspondingly decreases; the increase threshold θ and the decrease threshold
Figure FDA0003401676500000028
Are all set according to experience;
for the preference set by the user, if the factor J is lower than the reduction threshold value continuously in a plurality of statistical periods
Figure FDA0003401676500000029
If the preference type is 01 type, suggesting the user to delete the preference, and if the preference type is segmented or continuous type, suggesting the user to modify the preference; for the preference which is not set by the user, if the judgment factor J is larger than the increase threshold theta in a plurality of continuous statistical periods, recommending the user to add the preference which is not set; the statistical period duration is set empirically.
4. The intelligent recommendation system based on user preferences as claimed in claim 1, wherein the keyword query is a query for store information similar to a given keyword; inquiring the geographic position, namely inquiring shop information in a certain distance range of a given position, wherein the certain distance range is set according to experience;
the query grouping module employs the following processing strategy:
step 1, dividing the query request into two types: a keyword query request and a geographic position query request;
step 2, processing the query request according to the type of the query request:
(A) keyword-oriented query grouping:
step A1, presetting a similarity threshold, calculating the similarity between two keyword requests, if the similarity is greater than or equal to the preset threshold, judging that the two keyword requests are similar, and dividing the similar requests into the same group; the similarity calculation formula between any two keyword requests is as follows:
Figure FDA0003401676500000031
wherein q is1A keyword request indicating that the number of keywords is smaller than the number of keywords in the other keyword request among the two keyword requests; k represents q1Arbitrary key of (1), q1Key represents q1Key word of (1) | q1Key | represents q1Number of keywords in (1), N (k) represents q2Whether there is a synonym for k in the key;
the formula for N (k) is as follows:
Figure FDA0003401676500000032
step A2, scoring the query packet by using a set scoring function f (Q), wherein f (Q) is calculated as follows:
Figure FDA0003401676500000033
where | Qg | represents the number of packets for all requests, QiRepresenting an arbitrary packet request, Sim (Q)i) Representing a packet QiThe calculation formula is as follows:
Figure FDA0003401676500000034
wherein, | QiI represents a packet QiThe number of requests;
a3, if the score of the query group is larger than the threshold value, the group is qualified, and the qualified query group is sent to the query processing module for processing; if the score of the query grouping is smaller than the threshold value, the grouping is unqualified, and the grouping is deleted; the threshold value is set according to experience;
(B) geographic location-oriented query grouping:
step B1, presetting a similarity value threshold, calculating the similarity between any two geographical position query requests, if the similarity is less than or equal to the preset value, judging that the two geographical position query requests are similar, and dividing the similar geographical position query requests into the same group, wherein the calculation formula of the similarity between any two geographical position query requests is as follows:
Figure FDA0003401676500000041
wherein q is1Pos represents q1Position information of (a), q2Pos represents q2The location information of (a); if Sim (q)1，q2) If the geographic position query requests are 1, judging that the two geographic position query requests are similar, and dividing the similar geographic position query requests into the same group; if Sim (q)1，q2) If the two geographic position query requests are not similar to each other, judging that the two geographic position query requests are not similar to each other; the set distance is set according to experience;
step B2, scoring the packet by using a scoring function f (Q);
step B3, if the score of the query group is larger than the threshold value, the group is qualified, and the qualified query group is sent to the query processing module for processing; if the score of the query grouping is smaller than the threshold value, the grouping is unqualified, and the grouping is deleted; the threshold is set empirically.
5. The intelligent recommendation system based on user preferences as claimed in claim 1, wherein the query processing module is based on the following processing strategies:
step 1, retrieving the groups sent by the query grouping module, acquiring shops meeting the query keywords, and adding the acquired shops into a query result set:
step 1A, keyword-oriented query is based on the following processing strategies:
step a1, selecting a main query for a plurality of keyword queries in the group, wherein the main query is the query with the highest similarity score with other queries in the group, and the query similarity score calculation formula is as follows:
Figure FDA0003401676500000042
wherein Q isiRepresenting the ith group of keyword queries, qiIs QiThe ith keyword query of the group, Sim (q)iQ) is qiSimilarity to other keyword queries in the group;
step a2, processing the main query: if the keywords contained in the main query completely cover the keywords contained in other queries in the group, returning the result of the main query to other queries in the group; otherwise, merging the obtained main query and other queries which are not contained into a new query, and repeating the step a1 until a main query meeting the convergence condition is obtained, wherein the convergence condition is as follows: the keywords contained in the main query completely cover the keywords contained in other queries in the group;
step a3, retrieving keywords in the main query to obtain shops meeting the query keywords, and adding the obtained shops into a query result set;
step 1B, querying facing to the geographic position, based on the following processing strategies:
b1, merging the query ranges of the plurality of query conditions in the group, scanning the merged query ranges, and extracting all shop results in the ranges;
b2, judging the extracted results of each shop, judging whether the shop meets the query requirement, if so, adding the results into a query result set, wherein the query requirement is set according to experience;
step b3, sending the stores with the concentrated query results to a preference acquisition module, and calculating the store scores;
b4, sorting the query result set based on the user preference: and sequencing the query results according to the shop scores calculated by the preference acquisition module to obtain a top-k result set R, selecting the preferred skyline recommendation from the top-k result set R as a recommendation result, and sending the recommendation result to the user, wherein the top-k result set R is the query result selected according to requirements, and the requirements are set according to experience.
6. The intelligent recommendation system based on user preferences of claim 5,
the selection of the preferred skyline is based on the following processing strategy:
for each query result in the top-k result set R, performing skyline filtering based on each preference attribute dimension of the query result, wherein the skyline filtering comprises the following steps: for any two stores H in R1、H2If the shop H1Store H is not less than or equal to each preference attribute2At least one preference attribute is greater than store H2Then call H1Dominating H2(ii) a The dominated stores in the top-k result set R are deleted, and the remaining stores, as preferred skyline points, are returned to the user.
7. The intelligent recommendation system based on user preferences as claimed in claim 1, wherein the query keyword inputted by the user is used as basic preference information for the first time, and the background preference information is supplemented according to the group characteristics of the user.
CN202111498122.4A 2021-12-09 2021-12-09 Intelligent recommendation system based on user preference Pending CN114140170A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111498122.4A CN114140170A (en) 2021-12-09 2021-12-09 Intelligent recommendation system based on user preference

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111498122.4A CN114140170A (en) 2021-12-09 2021-12-09 Intelligent recommendation system based on user preference

Publications (1)

Publication Number Publication Date
CN114140170A true CN114140170A (en) 2022-03-04

Family

ID=80385177

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111498122.4A Pending CN114140170A (en) 2021-12-09 2021-12-09 Intelligent recommendation system based on user preference

Country Status (1)

Country Link
CN (1) CN114140170A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116662654A (en) * 2023-05-26 2023-08-29 惠州市西子湖畔网络有限公司 Big data-based affinity matching system and method

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116662654A (en) * 2023-05-26 2023-08-29 惠州市西子湖畔网络有限公司 Big data-based affinity matching system and method
CN116662654B (en) * 2023-05-26 2024-04-09 惠州市西子湖畔网络有限公司 Big data-based affinity matching system and method

Similar Documents

Publication Publication Date Title
US11238120B1 (en) Selecting, ranking, and/or presenting microsite content
CN103714130B (en) Video recommendation system and method
JP4426563B2 (en) Information provision system
CN102142003B (en) Method and device for providing point of interest information
CN107180093B (en) Information searching method and device and timeliness query word identification method and device
US20110145234A1 (en) Search method and system
WO2019056661A1 (en) Search term pushing method and device, and terminal
CN107657347B (en) Group-oriented travel recommendation method
CN105574216A (en) Personalized recommendation method and system based on probability model and user behavior analysis
CN102184224A (en) System and method for screening search results
WO2014176241A2 (en) Explanations for personalized recommendations
WO2014176192A2 (en) Live recommendation generation
CN110717093B (en) Movie recommendation system and method based on Spark
JP5831204B2 (en) Information providing system, information providing method, and program
KR102227552B1 (en) System for providing context awareness algorithm based restaurant sorting personalized service using review category
CN106446157B (en) Travel destination recommendation method and device
CN108446351B (en) Hotel screening method and system based on user preference of OTA platform
US20070143255A1 (en) Method and system for delivering internet content to mobile devices
CN107682178A (en) A kind of mobile subscriber's online operation behavior Forecasting Methodology and device
CN114140170A (en) Intelligent recommendation system based on user preference
KR102284393B1 (en) Apparatus and method for recommending learning contents
US20040024748A1 (en) Optimization of server selection using euclidean analysis of search terms
JP5723835B2 (en) Interest analysis method, interest analysis apparatus and program thereof
CN111078997B (en) Information recommendation method and device
CN112380451A (en) Favorite content recommendation method based on big data

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