CN106910104A - Application of the negative sequence pattern based on individual event missing in commercial product recommending - Google Patents
Application of the negative sequence pattern based on individual event missing in commercial product recommending Download PDFInfo
- Publication number
- CN106910104A CN106910104A CN201710018624.XA CN201710018624A CN106910104A CN 106910104 A CN106910104 A CN 106910104A CN 201710018624 A CN201710018624 A CN 201710018624A CN 106910104 A CN106910104 A CN 106910104A
- Authority
- CN
- China
- Prior art keywords
- negative
- sequence
- positive
- commodity
- pattern
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- 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
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
- G06Q30/0631—Item recommendations
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Game Theory and Decision Science (AREA)
- Data Mining & Analysis (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to be based on the single application for the negative sequence pattern of item not occurring in commercial product recommending, the present invention is applied during customers buying behavior analysis is carried out, the true buying behavior of customer can be fully represented, so as to find that customer really frequently buys or do not buy the behavior of commodity.So customer is when commodity are bought, can recommend some other client's purchase frequency than the product of larger related purchase to it using the present invention, so as to increase the Transaction apparatus meeting of customer, website browsing person is changed into buyer, improve cross-selling ability, the loyalty of client is improved, so as to improve the economic benefit of website.
Description
Technical field
The present invention relates to be based on application of the negative sequence pattern of individual event missing in commercial product recommending, belong to negative sequence pattern
Applied technical field.
Background technology
With the fast development of internet, shopping at network is from large size city to small and medium cities again to rural area gradually generation
Having replaced traditional shopping form turns into the dominant form of shopping.Shopping at network is built upon on the basis of mutual trusting and mutual benefiting, gram
The obstacle of region and time is taken, for consumer provides abundant merchandise news.Consumer can be multiple by browsing Duo Jia retail shops
Website, it might even be possible to which sea washes in a pan to obtain more price competitiveness, personalized, the commodity of customization.So brought to consumer
Bigger shopping, more commodity selections occupy big advantage in shopping.Therefore, shopping at network in recent years is in quick-fried
Hairdo increases, and is all increased with geometry several levels every year, at the same time many large-scale e-commerce websites, such as Amazon, Arriba
Taobao and day cat store, Jingdone district under bar etc. all have accumulated substantial amounts of customer transactional data.How these data are made full use of
Effective analysis, tissue are carried out to customers buying behavior to utilize, and how to recognize client's hobby as much as possible and value orientation,
It is that client provides personalized service to optimize website design, as the problem that e-commerce development compels highly necessary to solve.
Compared with traditional management style, the information contents of products of shopping at network is big, and quantity, species are enriched, and range of choice is wide.
Under traditional shopping environment, the source that consumer obtains merchandise news accumulates mainly by life, and collection process is more long, the letter for obtaining
Breath is more unilateral.Under shopping at network environment, consumer can just be collected with time of concentration, search the substantial amounts of letter for having a underlying commodity
Breath.Online type of merchandize is enriched, and some commodity consumptions person wants purchase and traditional retail shop is not easily found, it is possible to by network
Purchase is easily inquired about in shop, the shortage supplemented with some products of traditional store.But current e-commerce merchants are generally not
Can intuitively go to understand client, the related data of acquisition are limited (log-on message of such as user, purchaser record etc.).By right
Substantial amounts of client's purchaser record is analyzed and excavates, and finds the frequent access sequence pattern of client, belongs to for different clients
Property and shopping online step, using different commercial product recommending forms, the in good time commodity appropriate to lead referral, and optimize electronics
The putting position of business web site commodity, can effectively increase the Transaction apparatus meeting of client, and website browsing person is changed into buyer,
Cross-selling ability is improved, the loyalty of client, and the service quality and economic benefit for improving shopping website is improved.
Sequence pattern can not only analyze the relationship schedule between client trading commodity, can also analyze the pass between transaction
It is rule.So that seller is not only able to be had to client according to the situation of client trading commodity targetedly recommending, Er Qieneng
Later buying and selling of commodities situation is enough predicted according to current buying and selling of commodities situation such that it is able to preferably arrange the pendulum of commodity
Put.Whether its main purpose is to study the precedence relationship of commodity purchasing, finds out rule therein, i.e., do not only need to know commodity
It is purchased, and it needs to be determined that the sequencing of the commodity and other commodity purchasings, for example, a typical case of on-line purchase DVD
Order be purchase " Star War ", it is very possible afterwards to continue to buy " Caesarean counterattacks ", then be that " Jie Da warrior returns for purchase
Come ".Therefore sequence pattern it can be found that in database certain a period of time in Frequent episodes, i.e., which business within this time period
The comparing that product can be bought by client is more, more or few standard is determined by minimum support.Each sequence is according to transaction
One of Time alignment set, the sequence that minimum support meets different frequent degrees to excavate can be set.But in application
Series pattern analysis customers buying behavior, when solving the problems, such as individual commodity recommendation, they only account for the event for having occurred,
Referred to as positive sequence pattern (Positive Sequential Pattern, PSP) is excavated.
But with going deep into for research, researcher has found to imply substantial amounts of useful information in not generation event, and this
A little information are at all unavailable in simple positive sequence mode excavation.Then researcher will not occur again event consider exist
It is interior, it is proposed that negative sequence mode excavation.Negative sequence pattern (Negative Sequential Pattern, NSP) is excavated and not only examined
The event having occurred and that is considered, has been also directed to event, what it can analyze deeper into ground and understand in data potential has contained
Justice.For example, a represents bread, b represents coffee, and c represents tea,Represent client's purchase sequence pattern, the mode declaration pattern specification
Within certain a period of time, the client after it have purchased commodity a, in the case of without purchase commodity b, have purchased commodity c without
It is other commodity.Nowadays the value of negative sequence pattern increasingly approved by people, in intelligent checking system and much information
Application field plays irreplaceable effect, especially the buying behavior analysis of client is applied on commercial product recommending, even more
To commodity Successful Transaction and dealing money serves huge impetus.
Website user's purchase order data in e-commerce platform are the data source excavated.With 5 clients in 2 months
Transaction as a example by, if table 1 is the transaction database that is sorted by keyword by Customer ID and exchange hour.One Transaction Information
Storehouse a, affairs represent a transaction, and an individual event represents the commodity of transaction, and the letter record in individual event attribute is commodity
ID。
Table 1
Data prediction is carried out, the transaction database of table 1 is organized into the sequence library of table 2.
Table 2
Customer ID | Client's purchase sequence |
1 | <{c}{i}> |
2 | <{a,b}{c}{a,d,f,g}> |
3 | <{c,e,g,h}> |
4 | <{c}{c,d,g,h}{i}> |
5 | <{i}> |
One client's all of transaction record within certain time period constitutes an orderly sequence, and sequence is used<>Represent.
In the sequence, item/item collection is sequential, and each Xiang Dou represents a kind of commodity of transaction, and element refers to then the client at certain
One all commodity of disposable purchase of specific time point, represents, the client may purchase in the different time periods with { } or ()
Same part product is bought, i.e., one item may occur in a different elements for sequence.As ID is 2 client's purchase sequence in table 2
It is classified as<{a,b}{c}{a,d,f,g}>, the client respectively for the first time and third time do shopping when have purchased commodity a, wherein a,
B }, { c }, { a, d, f, g } these three Item Sets can be described as the element of sequence, and a, b, c, d, f, g are then referred to as item, if an element
Middle only one of which, then bracket can be omitted, and such as the element { c } in the sequence can directly write c.
However, existing negative sequence pattern mining algorithm is fewer, for example, NSPM, PNSP, Neg-GSP, e-NSP etc..
Most of constraints in these algorithms is proposed to realizing negative sequence mode excavation and improve efficiency of algorithm.It is true
On, these constraintss are relatively stricter so that many significant sequence patterns cannot be mined.They are in actual applications
It is not necessary to.For example, negative element is constrained, the minimal negative unit that it requires to bear candidate is element, has both allowed whole element
(ab) it is negative, is expressed asWithout allowing have negative term in element (ab), i.e.,Do not allow, this is very tight
Lattice, the commodity that the expression customer that it cannot be real has bought in once doing shopping, and the commodity that may be bought but not buy;Form is about
Beam, it requires that the negative element in sequence can not continuously occur, bothCan not occur, and in actual life
In this sequence be recurrent, it can represent the record of shopping twice of customer, and first time customer have purchased commodity b's
A is not bought simultaneously, second customer does not buy b while have purchased commodity c.But we are any currently without finding
It is related to relax the research of the negative mode method for digging of these constraintss.Because can be brought after relaxing constraint much having
How the problem of challenge, for example, producing the negative candidate of enormous amount, provide increasingly complex bearing and calculated comprising definition and height
Spend etc..
The content of the invention
Summary of the invention
The present invention excessively strictly causes the loss in mining process many significant for constraints in existing algorithm
Negative sequence pattern, and cannot really represent the buying behavior of customer these problems, it is proposed that the negative sequence based on individual event missing
Pattern mining algorithm NegI-NSP.The minimal negative unit that it not only allows to bear candidate is item (rather than element), for example, it is allowed to unit
There is negative term in plain (ab), can be expressed asOrAnd allow there is continuous negative element in sequence, for example, sequenceIt is allowed.So it is not only able to reflect the sequencing of commodity purchasing in customer purchasing behavior, also
The true buying behavior of customer in shopping every time can be reflected.
The main thought of NegI-NSP algorithms is:First, positive sequence is excavated by existing positive sequence pattern mining algorithm
Row pattern;Then, using converting the negative candidate sequence collection of discourse on politics generation, and support using known positive sequence pattern calculates negative
The support of candidate;Finally, screened by minimum support (being specified by user) and obtain final negative sequence pattern.
The negative sequence pattern that more can really reflect customer purchasing behavior for obtaining is excavated by the algorithm, so that can
More fully to analyze customers buying behavior so that seller can predict later business according to current merchandise sales situation
Product are sold, it is also possible to find out in shopping again customer's purchase well or do not buy the rule of commodity, so that reasonable arrangement business
Product are put, and improve offtake.
Detailed description of the invention
Technical scheme is as follows:
Application of the negative sequence pattern based on individual event missing in commercial product recommending, including step is as follows:
(1) define a data sequence and include a negative sequence
Customer will buy originally but the commodity without purchase are negative term, and behavior does not occur also referred to as;The once shopping of customer
It is recorded as element;Element comprising at least one negative term is negative element;Shopping record of the customer within a period of time is referred to as to count
According to sequence, the sequence comprising at least one negative element is referred to as negative sequence;
MPSE refers to a customer positive daughter element of the maximum in once shopping record;For example, a negative elementWhereinWithCustomer's specifically commodity of the shopping without purchase are represented, and b and d represent customer and specifically do shopping
The commodity that have purchased.Its maximum positron element representation isParticularly, individual event element is born not comprising just
, we are represented with sky, for example negative elementIts maximum positron element representation is
1-negMPSE refers to negative element siDaughter element, the daughter element include MPSE (si) and a negative term e,One
The 1-negMPSE of individual individual event negative element is itself;MPSE(si) refer to negative element siThe positive daughter element of maximum;
MPS (ns) refers to the positive subsequence of maximum of the negative sequence ns that the commodity bought by customer are constituted, and it is by ns
Comprising all positve terms constituted according to former order;For example:One negative sequence Represent without purchase
Commodity, and b c d represent the commodity that have purchased.Its positive subsequence of maximum be MPS (ns)=<c d>, particularly, a positive sequence
The positive subsequence of maximum of row is itself;
NSI-MSE refers to the subsequence of negative sequence ns, and the subsequence includes individual event negative element siAnd MPSE (si+1);
Define data sequence ds=<d1 d2 ... dt>Comprising negative sequence ns=<s1 s2... sm>;Will take into full account
To each negative term s in nsiNot in the interval occurred in ds, and this interval is by siMPS (ns) is divided into left subsequence
Determined with right subsequence, both first final position d of left subsequencefseWith the whole starting position d of right subsequencelsbBetween area
Between.
IfThen
Negative sequence ns=<s1 s2... sm>It is to be made up of m element, wherein comprising n negative element.
(1) if m>2t+1, then ds is not comprising ns;
(2) m=1 is worked as, during n=1, ifThen ds does not include ns;
(3) forAndWherein EidS- nsRepresent the negative element in ns
Set, EidS+ nsThe set of the positive element in ns is represented, if met
A () is as i=1, and (s1,id(s1))∈EidS- ns,(s2,id(s2))∈EidS+ ns, ((lsb=1) or (lsb>
1), ifParticularly, element_length (s1(the element_length of)=1
(s1) represent s1Length), thenThen ds includes ns;
B () is as i=2, and (s1,id(s1))∈EidS- nsAnd element_length (s1)=1, (s2,id(s2))∈EidS- ns,
((lsb=1) or (lsb>1), if
WithAs (s1,id(s1))∈EidS- nsAnd element_length (s1)>1, or (s1,id(s1))∈EidS+ ns,
(s2,id(s2))∈EidS- ns, ((lsb=1) or (lsb>1) when, ifEspecially
, if element_length (s2)>1, thenThen ds includes ns;
(c) as i=m, (sm-1,id(sm-1))∈EidS+ nsOr (sm-1,id(sm-1))∈EidS- ns, element_
length(sm-1)>1,(sm,id(sm))∈EidS- ns, ((fse=t) or (0<fse<T), ifSpecial element_length (sm)>1, thenWhen
(sm-1,id(sm-1))∈EidS- ns∧element_length(sm-1)=1, (sm,id(sm))∈EidS- ns,(0<fse<T), such as
ReallyThen ds includes ns;
D () is when 2<i<During m, (si-1,id(si-1))∈EidS+ nsOr (si-1,id(si-1))∈EidS- ns, element_
length(si-1)>1, (si,id(si))∈EidS- ns, ((si+1,id(si+1))∈EidS+ nsOr element_length (si)>
1),(fse>0 ∧ lsb=fse+1) or (fse>0∧lsb>Fse+1), if
If element_length (si)>1, thenAs (si-1,id(si-1))∈EidS- nsAnd element_length
(si-1)=1, (si,id(si))∈EidS- ns,(fse>0 and lsb>Fse+1), ifThen ds includes ns;Wherein fse=FSE (MPS (<s1 s2 ... si-1>),
), ds lsb=LSB (MPS (<si...sm>),ds);
(2) Shopping Behaviors of client are analyzed using NegI-NSP algorithms, specific steps include:
A, using positive sequence mining algorithm GSP excavate obtain all of positive sequence pattern, i.e., certain a period of time in, client
The commodity frequently bought;
Candidate sequence (negative is born in B, the positive sequence pattern obtained based on step A excavations, generation accordingly
Sequential candidates, NSC), the negative candidate sequence is used to judge within certain a period of time, which commodity client
That buys is more, and which commodity client does not buy;
C, the support that negative candidate sequence is calculated using the support of the corresponding positive sequence pattern of negative candidate sequence;
The negative sequence pattern for meeting minimum support requirement is filtered out in D, the negative candidate sequence generated from step B, i.e.,:From
The negative sequence pattern of the minimum support set more than user is filtered out in the negative candidate sequence of step B generations, using these sieves
The buying behavior of negative sequence pattern after choosing to client is analyzed.Businessman has according to analysis result and targetedly pushes away to client
Commodity are recommended, the putting position and quantity of commodity are arranged according to client's buying habit.
According to currently preferred, the step B, the positive sequence pattern for obtaining is excavated based on step A, generation is corresponding negative
Candidate sequence, specific steps include:For comprising the k positive sequence pattern of element, its negative candidate sequence is by changing positive sequence
Any m negative element is obtained in row pattern;M is integer and 1≤m≤k, is 1 when not having continuous length in positive sequence pattern
Element when, then the maximum of m be k.
For example:It is based on<(abc)cd>Positive sequence schema creation bears candidate sequence accordingly, and a, b, c, d refer to certain commodity,
Including:
M=1,
M=2,
In order to be based on the irredundant negative candidate sequence of positive sequence schema creation and combine the true shopping of customer in reality
Behavior, we generate negative candidate sequence with a kind of efficient method, and its basic thought is to change Arbitrary Term in positive sequence pattern
It is negative term, if there is continuous individual event element in positive sequence pattern, simultaneously can not changes into continuous negative individual event element
Item in negative element, and any multinomial element can not entirely change into negative term.
According to currently preferred, the step C, negative candidate sequence is calculated using the support of related positive sequence pattern
The support of row, specific steps include:
1. a negative candidate sequence is defined;For example,
1-neglMSnsRefer to the subsequence of negative sequence ns, and the subsequence is by MPS (ns) and a negative term group
Into;
1-neglMSSnsRefer to all 1-neglMS in negative sequence nsnsThe arrangement set of composition;For example,
P (1-neglMS) refers to that the positive element in sequence 1-neglMS is constant, and negative element is converted into corresponding positive element;
For example:
2. one containing m element and wherein containing the n negative sequence ns of negative element, for
1≤i≤n, in sequence library D, the support sup (ns) of negative sequence ns is calculated by formula (i), formula (ii):
Shown in the computational methods such as formula (i) of the support sup (ns) of negative sequence ns:
Sup (ns)=| { MPS (ns) }-∪n I=1{p(1-neglMSi)}| (i)
The sid of all data sequences comprising negative sequence ns is obtained using formula (i), { MPS (ns) } refers to all comprising MPS
(ns) sid of data sequence, ∪n I=1{p(1-neglMSi) refer to all { p (1-neglMSi) composition sid unions of sets
Collection;
When negative sequence ns isWhen, shown in the computational methods such as formula (ii) of the support sup (ns) of negative sequence ns:
It is assumed that sequence<(ac)>Support be 10, support sequence<(abc)>Sid (include sequence<(abc)>Visitor
Family ID) set be { 10,20,30 }, support sequence<(ac)d>The set of sid be { 20,30,40 }, then
The pseudo-code of the algorithm for realizing algorithm NegI-NSP of the invention is as follows:
Input:D:Client's purchase sequence database;ms:It is minimum support;
Output:NSP:Set for analyzing the negative sequence pattern of customers buying behavior;
The step (1) is excavated from customer purchasing behavior sequence library with existing positive sequence pattern mining algorithm
Go out all of positive sequence pattern;The set of all of positive sequence pattern and its support and sid is all stored in Hash table
PSPHash, wherein, the step (2) is the Hash codes of positive sequence pattern as key;
The step (4) is for each positive sequence pattern, by above-mentioned " generation of negative candidate sequence " method next life
Into negative candidate sequence NSC;
Step (5) calculates the support of each nsc in NSC by formula (i) and (ii) to step (20);Step
Suddenly (21) to step (23) is to judge to meet the negative sequence pattern NSP of minimum support;
Step (5) calculates the branch for comprising only a nsc for individual event negative element to step (8) by formula formula (ii)
Degree of holding, for the support of the nsc comprising multinomial negative element or element number more than 1, is calculated such as step by formula (i)
(9) to step (20);For the latter, it is crucial that how to calculate cHash.size ();
Step (12) to step (18) is the core concept for calculating cHash.size ();Will be comprising p (1-neglMSi)
{ p (1-neglMS are arrived in sid set storagesi) set in, then calculate { p (1-neglMSi) union, then calculate set in
The number of the sid for containing;Step (19) row calculates the support of nsc;
If nsc.support >=ms (nsc) so nsc is added into NSP, such as step (21) to step (23);
Returning result, such as step (26), then screened the sequence pattern that can be used for decision-making with appropriate screening technique,
The buying behavior of client is analyzed using the sequence pattern after these screenings, helps businessman to carry out targetedly Recommendations,
Commodity are put on demand.
Beneficial effects of the present invention are:
The present invention is applied during customers buying behavior analysis is carried out, and can fully represent the true purchase row of customer
For so as to find that customer really frequently buys or do not buy the behavior of commodity.So customer buy commodity when, using this hair
Bright product that can be to its some other client's purchase frequency of recommendation than larger related purchase, so as to increase the Transaction apparatus of customer
Meeting, buyer is changed into by website browsing person, improves cross-selling ability, the loyalty of client is improved, so as to improve website
Economic benefit.
Specific embodiment
The present invention is further qualified with reference to embodiment, but not limited to this.
Embodiment
Application of the negative sequence pattern based on individual event missing in commercial product recommending, including step is as follows:
(1) define a data sequence and include a negative sequence
Customer will buy originally but the commodity without purchase are negative term, and behavior does not occur also referred to as;The once shopping of customer
It is recorded as element;Element comprising at least one negative term is negative element;Shopping record of the customer within a period of time is referred to as to count
According to sequence, the sequence comprising at least one negative element is referred to as negative sequence;
MPSE refers to a customer positive daughter element of the maximum in once shopping record;For example, a negative elementWhereinWithCustomer's specifically commodity of the shopping without purchase are represented, and b and d represent customer and specifically do shopping
The commodity that have purchased.Its maximum positron element representation isParticularly, individual event element is born not comprising just
, we are represented with sky, for example negative elementIts maximum positron element representation is
1-negMPSE refers to negative element siDaughter element, the daughter element include MPSE (si) and a negative term e,
The 1-negMPSE of one individual event negative element is itself;MPSE(si) refer to negative element siThe positive daughter element of maximum;
MPS (ns) refers to the positive subsequence of maximum of the negative sequence ns that the commodity bought by customer are constituted, and it is by ns
Comprising all positve terms constituted according to former order;For example:One negative sequence Represent without purchase
Commodity, and b c d represent the commodity that have purchased.Its positive subsequence of maximum be MPS (ns)=<c d>, particularly, a positive sequence
The positive subsequence of maximum of row is itself;
NSI-MSE refers to the subsequence of negative sequence ns, and the subsequence includes individual event negative element siAnd MPSE (si+1);
Define data sequence ds=<d1 d2 ... dt>Comprising negative sequence ns=<s1 s2... sm>;Will take into full account
To each negative term s in nsiNot in the interval occurred in ds, and this interval is by siMPS (ns) is divided into left subsequence
Determined with right subsequence, both first final position d of left subsequencefseWith the whole starting position d of right subsequencelsbBetween area
Between.
IfThen
Negative sequence ns=<s1 s2... sm>It is to be made up of m element, wherein comprising n negative element.
(1) if m>2t+1, then ds is not comprising ns;
(2) m=1 is worked as, during n=1, ifThen ds does not include ns;
(3) forAndWherein EidS- nsRepresent the negative element in ns
Set, EidS+ nsThe set of the positive element in ns is represented, if met
A () is as i=1, and (s1,id(s1))∈EidS- ns,(s2,id(s2))∈EidS+ ns, ((lsb=1) or (lsb>
1), ifParticularly, element_length (s1(the element_length of)=1
(s1) represent s1Length), thenThen ds includes ns;
B () is as i=2, and (s1,id(s1))∈EidS- nsAnd element_length (s1)=1, (s2,id(s2))∈EidS- ns,
((lsb=1) or (lsb>1), if
WithAs (s1,id(s1))∈EidS- nsAnd element_length (s1)>1, or (s1,id(s1))∈EidS+ ns,
(s2,id(s2))∈EidS- ns, ((lsb=1) or (lsb>1) when, ifEspecially
, if element_length (s2)>1, thenThen ds includes ns;
(c) as i=m, (sm-1,id(sm-1))∈EidS+ nsOr (sm-1,id(sm-1))∈EidS- ns, element_
length(sm-1)>1,(sm,id(sm))∈EidS- ns, ((fse=t) or (0<fse<T), ifSpecial element_length (sm)>1, thenWhen
(sm-1,id(sm-1))∈EidS- ns∧element_length(sm-1)=1, (sm,id(sm))∈EidS- ns,(0<fse<T), such as
ReallyThen ds includes ns;
D () is when 2<i<During m, (si-1,id(si-1))∈EidS+ nsOr (si-1,id(si-1))∈EidS- ns, element_
length(si-1)>1, (si,id(si))∈EidS- ns, ((si+1,id(si+1))∈EidS+ nsOr element_length (si)>
1),(fse>0 ∧ lsb=fse+1) or (fse>0∧lsb>Fse+1), if
If element_length (si)>1, thenAs (si-1,id(si-1))∈EidS- nsAnd element_length
(si-1)=1, (si,id(si))∈EidS- ns,(fse>0 and lsb>Fse+1), ifThen ds includes ns;Wherein fse=FSE (MPS (<s1 s2 ... si-1>),ds),
Lsb=LSB (MPS (<si ... sm>),ds);
(2) Shopping Behaviors of client are analyzed using NegI-NSP algorithms, specific steps include:
A, using positive sequence mining algorithm GSP excavate obtain all of positive sequence pattern, i.e., certain a period of time in, client
The commodity frequently bought;
Candidate sequence (negative is born in B, the positive sequence pattern obtained based on step A excavations, generation accordingly
Sequential candidates, NSC), the negative candidate sequence is used to judge within certain a period of time, which commodity client
That buys is more, and which commodity client does not buy;Specific steps include:For comprising the k positive sequence pattern of element, its negative marquis
Any m negative element is obtained during sequence is selected by changing positive sequence pattern;M is integer and 1≤m≤k, when positive sequence pattern
In when there is no the element that continuous length is 1, then the maximum of m is k.
For example:It is based on<(abc)cd>Positive sequence schema creation bears candidate sequence accordingly, and a, b, c, d refer to certain commodity,
Including:
M=1,
M=2,
In order to be based on the irredundant negative candidate sequence of positive sequence schema creation and combine the true shopping of customer in reality
Behavior, we generate negative candidate sequence with a kind of efficient method, and its basic thought is to change Arbitrary Term in positive sequence pattern
It is negative term, if there is continuous individual event element in positive sequence pattern, simultaneously can not changes into continuous negative individual event element
Item in negative element, and any multinomial element can not entirely change into negative term.
C, the support that negative candidate sequence is calculated using the support of the corresponding positive sequence pattern of negative candidate sequence;Specifically
Step includes:
1. a negative candidate sequence is defined;For example,
1-neglMSnsRefer to the subsequence of negative sequence ns, and the subsequence is by MPS (ns) and a negative term group
Into;
1-neglMSSnsRefer to all 1-neglMS in negative sequence nsnsThe arrangement set of composition;For example,
P (1-neglMS) refers to that the positive element in sequence 1-neglMS is constant, and negative element is converted into corresponding positive element;
For example:
2. one containing m element and wherein containing the n negative sequence ns of negative element, for
1≤i≤n, in sequence library D, the support sup (ns) of negative sequence ns is calculated by formula (i), formula (ii):
Shown in the computational methods such as formula (i) of the support sup (ns) of negative sequence ns:
Sup (ns)=| { MPS (ns) }-∪n I=1{p(1-neglMSi)}| (i)
The sid of all data sequences comprising negative sequence ns is obtained using formula (i), { MPS (ns) } refers to all comprising MPS
(ns) sid of data sequence, ∪n I=1{p(1-neglMSi) refer to all { p (1-neglMSi) composition sid unions of sets
Collection;
When negative sequence ns isWhen, shown in the computational methods such as formula (ii) of the support sup (ns) of negative sequence ns:
It is assumed that sequence<(ac)>Support be 10, support sequence<(abc)>Sid (include sequence<(abc)>Visitor
Family ID) set be { 10,20,30 }, support sequence<(ac)d>The set of sid be { 20,30,40 }, then
The negative sequence pattern for meeting minimum support requirement is filtered out in D, the negative candidate sequence generated from step B, i.e.,:From
The negative sequence pattern of the minimum support set more than user is filtered out in the negative candidate sequence of step B generations, using these sieves
The buying behavior of negative sequence pattern after choosing to client is analyzed.Businessman has according to analysis result and targetedly pushes away to client
Commodity are recommended, the putting position and quantity of commodity are arranged according to client's buying habit.
The pseudo-code of the algorithm for realizing algorithm NegI-NSP of the invention is as follows:
Input:D:Client's purchase sequence database;ms:It is minimum support;
Output:NSP:Set for analyzing the negative sequence pattern of customers buying behavior;
The step (1) is excavated from customer purchasing behavior sequence library with existing positive sequence pattern mining algorithm
Go out all of positive sequence pattern;The set of all of positive sequence pattern and its support and sid is all stored in Hash table
PSPHash, wherein, the step (2) is the Hash codes of positive sequence pattern as key;
The step (4) is for each positive sequence pattern, by above-mentioned " generation of negative candidate sequence " method next life
Into negative candidate sequence NSC;
Step (5) calculates the support of each nsc in NSC by formula (i) and (ii) to step (20);Step
Suddenly (21) to step (23) is to judge to meet the negative sequence pattern NSP of minimum support;
Step (5) calculates the branch for comprising only a nsc for individual event negative element to step (8) by formula formula (ii)
Degree of holding, for the support of the nsc comprising multinomial negative element or element number more than 1, is calculated such as step by formula (i)
(9) to step (20);For the latter, it is crucial that how to calculate cHash.size ();
Step (12) to step (18) is the core concept for calculating cHash.size ();Will be comprising p (1-neglMSi)
{ p (1-neglMS are arrived in sid set storagesi) set in, then calculate { p (1-neglMSi) union, then calculate set in
The number of the sid for containing;Step (19) row calculates the support of nsc;
If nsc.support >=ms (nsc) so nsc is added into NSP, such as step (21) to step (23);
Returning result, such as step (26), then screened the sequence pattern that can be used for decision-making with appropriate screening technique,
The buying behavior of client is analyzed using the sequence pattern after these screenings, helps businessman to carry out targetedly Recommendations,
Commodity are put on demand.
Claims (3)
1. application of the negative sequence pattern based on individual event missing in commercial product recommending, it is characterised in that as follows including step:
(1) define a data sequence and include a negative sequence
Customer will buy originally but the commodity without purchase are negative term, and behavior does not occur also referred to as;The once shopping record of customer
It is element;Element comprising at least one negative term is negative element;Shopping record of the customer within a period of time is referred to as data sequence
Row, the sequence comprising at least one negative element is referred to as negative sequence;
MPSE refers to a customer positive daughter element of the maximum in once shopping record;
1-negMPSE refers to negative element siDaughter element, the daughter element include MPSE (si) and a negative term e,One list
The 1-negMPSE of item negative element is itself;MPSE(si) refer to negative element siThe positive daughter element of maximum;
MPS (ns) refers to the positive subsequence of maximum of the negative sequence ns that the commodity bought by customer are constituted, and it in ns by including
All positve terms constituted according to former order;
NSI-MSE refers to the subsequence of negative sequence ns, and the subsequence includes individual event negative element siAnd MPSE (si+1);
Define data sequence ds=<d1 d2 ... dt>Comprising negative sequence ns=<s1 s2 ... sm>;
(2) Shopping Behaviors of client are analyzed using NegI-NSP algorithms, specific steps include:
A, using positive sequence mining algorithm GSP excavate obtain all of positive sequence pattern, i.e., certain a period of time in, client is frequent
The commodity of purchase;
Candidate sequence is born in B, the positive sequence pattern obtained based on step A excavations, generation accordingly, and the negative candidate sequence is used to sentence
Break within certain a period of time, it is many which commodity client buys, and which commodity client does not buy;
C, the support that negative candidate sequence is calculated using the support of the corresponding positive sequence pattern of negative candidate sequence;
The negative sequence pattern for meeting minimum support requirement is filtered out in D, the negative candidate sequence generated from step B, i.e.,:From step
The negative sequence pattern of the minimum support set more than user is filtered out in the negative candidate sequence of B generations, after these screenings
Buying behavior of the negative sequence pattern to client be analyzed.
2. application of the negative sequence pattern based on individual event missing according to claim 1 in commercial product recommending, its feature exists
In the step B excavates the positive sequence pattern for obtaining based on step A, and candidate sequence is born in generation accordingly, and specific steps include:
For comprising the k positive sequence pattern of element, its negative candidate sequence be by changing positive sequence pattern in any m negative element obtain
Arrive;M is integer and 1≤m≤k, and when not having element that continuous length is 1 in positive sequence pattern, then the maximum of m is k.
3. application of the negative sequence pattern based on individual event missing according to claim 1 in commercial product recommending, its feature exists
In the step C calculates the support of negative candidate sequence, specific steps bag using the support of related positive sequence pattern
Include:
1. a negative candidate sequence is defined;
1-neglMSnsRefer to the subsequence of negative sequence ns, and the subsequence is made up of MPS (ns) and a negative term;
1-neglMSSnsRefer to all 1-neglMS in negative sequence nsnsThe arrangement set of composition;
2. one containing m element and wherein containing the n negative sequence ns of negative element, for1
≤ i≤n, in sequence library D, the support sup (ns) of negative sequence ns is calculated by formula (i), formula (ii):
Shown in the computational methods such as formula (i) of the support sup (ns) of negative sequence ns:
Sup (ns)=| { MPS (ns) }-∪n I=1{p(1-neglMSi)}| (i)
The sid of all data sequences comprising negative sequence ns is obtained using formula (i), { MPS (ns) } refers to all comprising MPS (ns)
Data sequence sid, ∪n I=1{p(1-neglMSi) refer to all { p (1-neglMSi) composition sid union of sets collection;
When negative sequence ns isWhen, shown in the computational methods such as formula (ii) of the support sup (ns) of negative sequence ns:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710018624.XA CN106910104A (en) | 2017-01-11 | 2017-01-11 | Application of the negative sequence pattern based on individual event missing in commercial product recommending |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710018624.XA CN106910104A (en) | 2017-01-11 | 2017-01-11 | Application of the negative sequence pattern based on individual event missing in commercial product recommending |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106910104A true CN106910104A (en) | 2017-06-30 |
Family
ID=59207315
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710018624.XA Pending CN106910104A (en) | 2017-01-11 | 2017-01-11 | Application of the negative sequence pattern based on individual event missing in commercial product recommending |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106910104A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111949711A (en) * | 2020-08-18 | 2020-11-17 | 齐鲁工业大学 | Commodity recommendation system based on decision-making high-utility negative sequence rule mining and working method thereof |
-
2017
- 2017-01-11 CN CN201710018624.XA patent/CN106910104A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111949711A (en) * | 2020-08-18 | 2020-11-17 | 齐鲁工业大学 | Commodity recommendation system based on decision-making high-utility negative sequence rule mining and working method thereof |
CN111949711B (en) * | 2020-08-18 | 2021-06-01 | 齐鲁工业大学 | Commodity recommendation system based on decision-making high-utility negative sequence rule mining and working method thereof |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kirby-Hawkins et al. | An investigation into the geography of corporate e-commerce sales in the UK grocery market | |
CN104504159A (en) | Application of multi-supporting-degree positive and negative sequence modes in clients' purchasing behavior analysis | |
CN104574153A (en) | Method for quickly applying negative sequence mining patterns to customer purchasing behavior analysis | |
CN109934369A (en) | Method and device for information push | |
KR20150053256A (en) | Customer data stochastic analysis system | |
CN104537553A (en) | Application of repeated negative sequence pattern in customer purchase behavior analysis | |
JP5914549B2 (en) | Information processing apparatus and information analysis method | |
US20160203138A1 (en) | Systems and methods for generating analytics relating to entities | |
Hu et al. | Tourism demand forecasting using nonadditive forecast combinations | |
Alatrista-Salas et al. | Impact of natural disasters on consumer behavior: Case of the 2017 El Niño phenomenon in Peru | |
Mostafa | Knowledge discovery of hidden consumer purchase behaviour: a market basket analysis | |
JP2016212792A (en) | Commodity recommendation device, commodity recommendation method and program | |
CN111127074A (en) | Data recommendation method | |
Uncles et al. | Generalizing patterns of store-type patronage: an analysis across major Chinese cities | |
Johansson et al. | Reputational assets and social media marketing activeness: Empirical insights from China | |
KR101479379B1 (en) | Method of sharing customer rate of merchandise based on social network | |
Kumar et al. | Cuisine prediction based on ingredients using tree boosting algorithms | |
CN106910104A (en) | Application of the negative sequence pattern based on individual event missing in commercial product recommending | |
Pramana et al. | Big data implementation for price statistics in Indonesia: Past, current, and future developments | |
Srivastava | Motif analysis in the amazon product co-purchasing network | |
Fitrianah et al. | Analysis of Consumer Purchase Patterns on Handphone Accessories Sales Using FP-Growth Algorithm | |
Chen et al. | Boosting recommendation in unexplored categories by user price preference | |
CN111460300B (en) | Network content pushing method, device and storage medium | |
Thill | Modeling store choices with cross-sectional and pooled cross-sectional data: a comparison | |
JP2016122472A (en) | Information processing apparatus and information analysis method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170630 |