CN103412882B - A kind of method and device identifying that consumption is intended to - Google Patents
A kind of method and device identifying that consumption is intended to Download PDFInfo
- Publication number
- CN103412882B CN103412882B CN201310301375.7A CN201310301375A CN103412882B CN 103412882 B CN103412882 B CN 103412882B CN 201310301375 A CN201310301375 A CN 201310301375A CN 103412882 B CN103412882 B CN 103412882B
- Authority
- CN
- China
- Prior art keywords
- daily record
- user
- intended
- consumption
- indication
- 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.)
- Active
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a kind of method and device identifying that consumption is intended to, set up the process of consumption intention assessment model: from the historical behavior daily record of each user, filter out the corelation behaviour daily record setting consumer field；The BMAT carried out based on corelation behaviour daily record, determine corresponding buy move ahead into user behaviors log and corresponding buy after the user behaviors log of behavior；Select to meet the user behaviors log of training data screening conditions as training sample from the user behaviors log determined；Therefrom extract features training disaggregated model, obtain setting the consumption intention assessment model that consumer field is corresponding；Identify the process that consumption is intended to: determine the consumer field of user to be identified；Utilizing the consumption intention assessment model that the consumer field that determines is corresponding, the corelation behaviour daily record to user to be identified in nearly a period of time is classified, after obtaining before the consumption of user to be identified is intended that purchase or buying.The present invention is capable of the identification that customer consumption is intended to, and contributes to carrying out information more accurately for user and throws in.
Description
[technical field]
The present invention relates to Computer Applied Technology field, particularly to a kind of identify consumption be intended to method and
Device.
[background technology]
Developing rapidly and gradually popularize along with network technology, occur in that carry out doing shopping by network, conclude the business,
The ecommerce new models such as popularization, ecommerce develops rapidly by its low cost, high efficiency advantage.
It addition, along with the raising day by day of people's material and cultural life, consumption demand also present variation,
At many levels, and progressively being developed to high-level by low level, consumer field constantly extends, and consumption content is day by day
Abundant trend.
Identification for customer consumption demand largely have impact on the such as information such as commodity, popularization input
Accuracy, wherein consumption be intended that a critically important factor.Such as, one is intended to buy hands
The user of machine, to the relevant goods purchase information of its recommending mobile phone, rate of exchange information, contrast evaluation and test information etc.
Meet user's request, but the user that have purchased mobile phone for, then to its recommending mobile phone phase
The goods purchase information of pass, rate of exchange information, contrast evaluation and test information etc. are the most suitable, now user's request
Be use the information such as gains in depth of comprehension information, relevant peripheral merchandise news, after sale information.But the most relatively
The excellent correlation technique for user view identification.
[summary of the invention]
In view of this, the invention provides a kind of method and device identifying that consumption is intended to, disappear in order to user
Take intention assessment, contribute to carrying out information more accurately for user and throw in.
Concrete technical scheme is as follows:
A kind of method identifying that consumption is intended to, the method includes:
The process of foundation consumption intention assessment model:
S11, filter out from the historical behavior daily record of each user set consumer field corelation behaviour daily record；
S12, based on the BMAT that the corelation behaviour daily record filtered out is carried out, before determining corresponding purchase
The user behaviors log of behavior after the user behaviors log of behavior and corresponding purchase；
S13, from the user behaviors log that step S12 determines, select to meet the user behaviors log of training data screening conditions
As training sample；
S14, from training sample, extract features training disaggregated model, obtain can recognize that purchase move ahead into
The consumption intention assessment model that after purchase, the described setting consumer field of behavior is corresponding；
The process that identification consumption is intended to:
S21, determine the consumer field of user to be identified；
S22, utilize consumption intention assessment model corresponding to the consumer field determined, to institute in nearly a period of time
State user to be identified to classify in the corelation behaviour daily record of the described consumer field determined, obtain to be identified
Before the consumption of user is intended that purchase or after purchase.
According to the present invention one preferred implementation, described step S11 specifically includes:
N the lists of keywords corresponding with described setting consumer field respectively by the historical behavior daily record of each user
Matching, filter out and comprise the user behaviors log of key word in described n lists of keywords simultaneously, described n is
Positive integer；Or,
The expression template that the historical behavior daily record of each user is corresponding with described setting consumer field respectively is matched,
Filter out the user behaviors log mated with described expression template.
According to the present invention one preferred implementation, the corelation behaviour daily record filtered out is entered by described step S12
The BMAT of row is:
Manually the corelation behaviour daily record filtered out is carried out BMAT；Or,
The key word being intended to according to the indication consumption comprised in corelation behaviour daily record carries out the analysis of behavioral pattern, institute
State indication consumption be intended to key word include indication buy before be intended to key word or indication buy after be intended to pass
Keyword.
According to the present invention one preferred implementation, described training data screening conditions include the one in following condition
Or combination in any:
The daily record quantity that the user behaviors log of user comprises is more than or equal to predetermined number threshold value；
The key word accounting that the indication consumption that user behaviors log is comprised is intended to is more than or equal to preset ratio threshold value, institute
State indication consumption be intended to key word be indication buy before be intended to key word or indication buy after be intended to key
Word；
User behaviors log occurring, what the occurrence number of most brands accounted for all brands in behavior daily record goes out occurrence
The ratio of number exceedes preset ratio threshold value.
According to the present invention one preferred implementation, the key word that described indication consumption is intended to digs in the following way
Pick:
A1, for described setting consumer field determine indication consumption be intended to seed words；
A2, utilize described seed words that the sample daily record of described setting consumer field is classified, bought
Sample daily record after front sample daily record and purchase；
A3, respectively to sample daily record before buying with after sample daily record carries out the filtration of participle and stop word after buying,
Add up the word frequency of each word；
A4, based on each word respectively sample daily record and the word frequency in sample daily record after buying the most before purchase, determine
The key word that the key word that indication is intended to before buying is intended to after buying with indication.
According to the present invention one preferred implementation, the feature extracted from training sample in described step S14
At least include but not limited to the one in following characteristics or combination in any:
Daily record quantity；
Daily record quantity shared by the brand that occurrence number is most；
The daily record quantity shared by brand that occurrence number time is many；
Daily record quantity shared by the model that occurrence number is most；
The daily record quantity shared by model that occurrence number time is many；
The key word accounting that indication consumption is intended to；
The appearance position of the key word that indication consumption is intended to；
There is the daily record accounting of query；
The key word accounting of the indication user view comprised in query；
Duration is crossed in daily record；
The maximum of daily record shared by same brand crosses over duration；
The maximum of daily record shared by same model crosses over duration.
According to the present invention one preferred implementation, described step S13 specifically includes: determine from step S12
User behaviors log selects to meet respectively N number of user behaviors log set of N group training data screening conditions；
Described step S14 specifically includes: each user behaviors log set instructed as one group of training sample
Practice disaggregated model, obtain N number of candidate family, from described N number of candidate family, select a conduct of optimum
The consumption intention assessment model that described setting consumer field is corresponding.
According to the present invention one preferred implementation, one of described optimum is: described N number of candidate family fullness in the epigastrium and abdomen
Foot is preset recall rate and is required and the highest one of accuracy rate.
According to the present invention one preferred implementation, after described step S22, also include:
S23, it is intended to corresponding exhibition strategy to described user's exhibition to be identified according to the consumption of described user to be identified
Show information.
According to the present invention one preferred implementation, described step S23 specifically includes:
It is intended to according to the consumption of described user to be identified, shows and described use to be identified to described user to be identified
The consumption at family is intended to the promotion message of corresponding types；Or,
In the Search Results that described user to be identified shows, it is right the consumption of described user to be identified to be intended to
The information answered sequence in Search Results is in advance.
A kind of device identifying that consumption is intended to, this device includes modeling unit and recognition unit；
Described modeling unit includes:
First screening subelement, sets consumer field for filtering out from the historical behavior daily record of each user
Corelation behaviour daily record；
Pattern analysis subelement, is used for based on the BMAT carrying out the corelation behaviour daily record filtered out,
Determine corresponding buy move ahead into user behaviors log and corresponding buy after the user behaviors log of behavior；
Second screening subelement, satisfied for selecting from the user behaviors log that described pattern analysis subelement determines
The user behaviors log of training data screening conditions is as training sample；
Model training subelement, for extracting features training disaggregated model from training sample, obtains knowing
Do not go out to buy move ahead into buy after consumption intention assessment model corresponding to the described setting consumer field of behavior；
Described recognition unit includes:
Field determines subelement, for determining the consumer field of user to be identified；
Category of model subelement is corresponding for the consumer field utilizing described field to determine that subelement is determined
Consumption intention assessment model, to user to be identified described in nearly a period of time in the described consumer field determined
Corelation behaviour daily record classify, after obtaining before the consumption of user to be identified is intended that purchase or buying.
According to the present invention one preferred implementation, described first screening subelement specifically performs:
N the lists of keywords corresponding with described setting consumer field respectively by the historical behavior daily record of each user
Matching, filter out and comprise the user behaviors log of key word in described n lists of keywords simultaneously, described n is
Positive integer；Or,
By expression template phase corresponding with described setting consumer field respectively for the historical behavior daily record of each user
Join, filter out the user behaviors log mated with described expression template.
According to the present invention one preferred implementation, described pattern analysis subelement based on artificial to the phase filtered out
Close the BMAT that user behaviors log is carried out, or, according to the indication consumption comprised in corelation behaviour daily record
The key word being intended to carries out the analysis of behavioral pattern, and the key word that described indication consumption is intended to includes that indication is bought
The key word that the key word of front intention or indication are intended to after buying.
According to the present invention one preferred implementation, described training data screening conditions include the one in following condition
Or combination in any:
The daily record quantity that the user behaviors log of user comprises is more than or equal to predetermined number threshold value；
The key word accounting that the indication consumption that user behaviors log is comprised is intended to is more than or equal to preset ratio threshold value, institute
State indication consumption be intended to key word be indication buy before be intended to key word or indication buy after be intended to key
Word；
User behaviors log occurring, what the occurrence number of most brands accounted for all brands in behavior daily record goes out occurrence
The ratio of number exceedes preset ratio threshold value.
According to the present invention one preferred implementation, described modeling unit also includes that being intended to key word excavates subelement,
For excavating the key word that described indication consumption is intended in such a way:
A1, for described setting consumer field determine indication consumption be intended to seed words；
A2, utilize described seed words that the sample daily record of described setting consumer field is classified, bought
Sample daily record after front sample daily record and purchase；
A3, respectively to sample daily record before buying with after sample daily record carries out the filtration of participle and stop word after buying,
Add up the word frequency of each word；
A4, based on each word respectively sample daily record and the word frequency in sample daily record after buying the most before purchase, determine
The key word that the key word that indication is intended to before buying is intended to after buying with indication.
According to the present invention one preferred implementation, described model training subelement when train classification models from instruction
Practicing the feature extracted in sample and described category of model subelement is utilizing consumption intention assessment model to carry out point
The feature extracted the corelation behaviour daily record of the described consumer field determined from described user to be identified during class
At least include but not limited to the one in following characteristics or combination in any:
Daily record quantity；
Daily record quantity shared by the brand that occurrence number is most；
The daily record quantity shared by brand that occurrence number time is many；
Daily record quantity shared by the model that occurrence number is most；
The daily record quantity shared by model that occurrence number time is many；
The key word accounting that indication consumption is intended to；
The appearance position of the key word that indication consumption is intended to；
There is the daily record accounting of query；
The key word accounting of the indication user view comprised in query；
Duration is crossed in daily record；
The maximum of daily record shared by same brand crosses over duration；
The maximum of daily record shared by same model crosses over duration.
According to the present invention one preferred implementation, described second screening subelement is from described pattern analysis sub-list
The user behaviors log that unit determines selects to meet respectively N number of user behaviors log collection of N group training data screening conditions
Close；
Described model training subelement is using each user behaviors log set as one group of training sample training point
Class model, obtains N number of candidate family, selects one of optimum as described from described N number of candidate family
Set the consumption intention assessment model that consumer field is corresponding.
According to the present invention one preferred implementation, one of described optimum is: described N number of candidate family fullness in the epigastrium and abdomen
Foot is preset recall rate and is required and the highest one of accuracy rate.
According to the present invention one preferred implementation, described recognition unit also includes: information shows subelement, uses
Letter is shown to described user to be identified in being intended to corresponding exhibition strategy according to the consumption of described user to be identified
Breath.
According to the present invention one preferred implementation, described information shows that subelement specifically performs:
It is intended to according to the consumption of described user to be identified, shows and described use to be identified to described user to be identified
The consumption at family is intended to the promotion message of corresponding types；Or,
In the Search Results that described user to be identified shows, it is right the consumption of described user to be identified to be intended to
The information answered sequence in Search Results is in advance.
As can be seen from the above technical solutions, the present invention can utilize user's historical behavior daily record to consume
The foundation of intent model, and utilize the consumption intent model of foundation to carry out the identification of customer consumption intention, i.e.
Identify user be buy before be intended to or buy after be intended to, thus contribute to carrying out for user more accurate
Information throw in.
[accompanying drawing explanation]
The method flow diagram setting up consumption intention assessment model that Fig. 1 provides for the embodiment of the present invention one；
The method for digging flow process of the key word that the indication consumption that Fig. 2 provides for the embodiment of the present invention one is intended to
Figure；
Fig. 3 identifies, for what the embodiment of the present invention two provided, the method flow diagram that consumption is intended to；
Fig. 4 identifies, for what the embodiment of the present invention three provided, the structure drawing of device that consumption is intended to.
[detailed description of the invention]
In order to make the object, technical solutions and advantages of the present invention clearer, below in conjunction with the accompanying drawings and specifically
Embodiment describes the present invention.
The present invention realizes, by being analyzed user's historical behavior daily record, the identification that consumption is intended to, main
Two parts process to be divided into: the first is consumed by being analyzed a large number of users historical behavior daily record setting up
The process of intention assessment model；Its two be utilize consumption intention assessment model user is carried out consumption be intended to know
Other process.By embodiment one and embodiment two, above-mentioned two process is described separately below.
Embodiment one,
The method flow diagram setting up consumption intention assessment model that Fig. 1 provides for the embodiment of the present invention one, this
The consumption intention assessment model set up in inventive embodiments is respectively directed to each different consumer fields, often
Individual consumer field has the consumption intention assessment model of its correspondence, the consumption of such as digital consumer product area to be intended to
Identify model, field of household appliances consumption intention assessment model, the consumption intention assessment model in house property field,
Consumption intention assessment model consuming intention assessment model, cosmetic field of automotive field etc..This reality
Execute example and describe the consumption identification method for establishing model of one of which consumer field, as it is shown in figure 1, the party
Method comprises the following steps:
Step 101: filter out the corelation behaviour setting consumer field from the historical behavior daily record of each user
Daily record.
This step actually filters out setting consumer field phase from the historical behavior daily record of a large number of users
The user behaviors log closed, thus the data source required for setting up as this consumer field subsequent consumption identification model.
Such as, when will set up the consumption intention assessment model of digital consumer product area, the history from each user is needed
User behaviors log filters out the corelation behaviour daily record of digital consumer product area.In order to ensure that customer consumption is intended to
Identifying that the screening of the real-time effect of model, generally this step is originated is to set each user in the time period recently
User behaviors log, the user behaviors log of user in the most nearest a week.
The mode of screening can use but the one that is not limited in the following manner or combination: based on key word
The mode joined, and mode based on template matching.
Mode based on Keywords matching is, excavates in advance and sets the key word row that consumer field is corresponding
Table, the lists of keywords that each user behaviors log is corresponding with setting consumer field respectively matches, and filters out bag
User behaviors log containing the key word in lists of keywords.Here lists of keywords can be one or more,
The user behaviors log of the key word comprised in any one lists of keywords can be filtered out, or filter out simultaneously
Comprise the user behaviors log of key word in several lists of keywords.
As a example by digital consumer product area, excavate the marque row that digital consumer product area is corresponding in advance
Table, such as " Nokia 5233 ", " HTC Desire ", " Nexus7 ", " ipad2 " etc.,
When performing this step and carrying out Log Filter, load this marque list, filter out and comprise this commodity-type
The user behaviors log of the key word in number list, such as comprise " Nokia 5233 " in certain daily record, then should
Log Filter is out.This is a kind of situation, also has a kind of situation: excavate digital consumer product area in advance
Corresponding list of brands and type of merchandise list, list of brands such as comprises " association ", " Fructus Mali pumilae ",
" Samsung ", " Nokia " etc., type of merchandise list comprises " notebook ", " mobile phone ",
" panel computer ", " camera " etc., filter out and comprise in list of brands and type of merchandise list simultaneously
The user behaviors log of key word, such as certain daily record comprises " Samsung " and " mobile phone " simultaneously, then should
Log Filter is out.
Mode based on template matching is, excavates in advance and sets the expression template that consumer field is corresponding, will
Each user behaviors log mates with this expression template respectively, filters out the behavior day mated with this expression template
Will.
Still as a example by digital consumer product area, excavate the expression mould that digital consumer product area is corresponding in advance
Plate, such as " display screen :+[digital]+pixel ", if a user behaviors log comprises " display screen:
8000000 pixels ", then by behavior Log Filter out.
Step 102: the BMAT carried out based on the corelation behaviour daily record filtered out, determines correspondence
Buy move ahead into user behaviors log and corresponding buy after the user behaviors log of behavior.
Generally, user before purchase, is generally in the preliminary analysis stage, it is of concern that price, parameter,
The problems such as performance, the information such as what user was concerned about after point of purchase is then after sale, use, periphery dependent merchandise,
Based on this characteristic, it is possible to after determining that user behaviors log correspondence is still bought before buying.BMAT
Can artificially be analyzed and each corelation behaviour daily record is marked analysis result, naturally it is also possible to using automatically
The mode analyzed.
If using the mode automatically analyzed, can be based on the indication comprised respectively in each corelation behaviour daily record
The key word that consumption is intended to carries out the analysis of behavioral pattern.Here, the key word that indication consumption is intended to also may be used
With referred to as polarity key word, the key that the key word being divided into indication to be intended to before buying is intended to after buying with indication
Word.Such as " offer ", " parameter ", the key word such as " price " indication are intended to before buying, " setting ",
The indication of the key word such as " maintenace point ", " software " is intended to after buying.The key that these indication consumption are intended to
Word can the most manually be arranged, but in order to improve efficiency, is conducive to extending new consumer goods, it is preferred to use
The mode of automatic mining.
The method for digging of the key word that indication consumption is intended to is as in figure 2 it is shown, comprise the following steps:
Step 201: for setting the seed words that consumer field determines that indication consumption is intended to.
This seed words can the most artificially be arranged in this step, it is common that for a certain concrete consumer field
Arrange, but a consumer field would generally comprise multiple subclass, when selected seed word, is preferably chosen at
The seed words that the indication being all suitable in each subclass is intended to after buying the front seed words being intended to and buying.Such as,
For digital consumer product area, generally include each subclasses such as computer, camera, mobile phone, can be by electricity at this
The key word being intended to before indication is bought in the subclasses such as brain, camera and mobile phone takes common factor and is used as digital consumer goods
The seed words being intended to before indication is bought in field.
Step 202: the sample daily record of this setting consumer field is entered by the seed words utilizing indication consumption to be intended to
Row classification, sample daily record after obtaining buying front sample daily record and buying.
Choose some sample daily records setting consumer field in advance, purchase if certain sample daily record comprises indication
The seed words that before buying, consumption is intended to meets pre-more than the situation comprising the seed words consuming intention after indication is bought
If condition, then it is assumed that this sample daily record is sample daily record before buying；If certain sample daily record comprises indication
The unnecessary situation comprising the seed words consuming intention before indication is bought of seed words consuming intention after purchase meets
Pre-conditioned, then it is assumed that this sample daily record is sample daily record after buying.Here pre-conditioned can use:
Difference is more than preset difference value threshold value, and difference accounts for the ratio of the seed words total quantity comprised in this sample daily record and surpasses
Cross preset ratio threshold value, etc..
Step 203: respectively sample daily record after sample daily record before buying and purchase is carried out participle and stop word
Filtration after, add up the word frequency of each word.
The stop word filtered can include but not limited to following at least one: relevant to brand, model
Word, the word insignificant to semantic meaning representation such as function word, auxiliary word, modal particle, have letter, numeral or one
The noise words such as individual Chinese character.
Step 204: based on the word in sample daily record after the sample daily record the most before purchase of each word and purchase
Frequently, the key word that the key word that indication is intended to before buying is intended to after buying is determined with indication.
If the word frequency in the sample daily record before purchase of certain word is the highest, but after point of purchase in sample daily record
Word frequency is the lowest or be 0, it is determined that this word is the key word being intended to before indication is bought, equally, if
Word frequency in the sample daily record after point of purchase of certain word is the highest, but the word frequency in sample daily record is the lowest before purchase
Or it is 0, it is determined that this word is the key word being intended to after indication is bought.
With continued reference to Fig. 1, step 103: select to meet training from the user behaviors log that step 102 determines
The user behaviors log of data screening condition is as training sample.
In view of the workload of artificial notation methods selection training sample is too big and inefficiency, can mark
Limited amount and dispute relatively big, use the side automatically selecting training sample the most in embodiments of the present invention
Formula.Above-mentioned training data screening conditions can use but be not limited to the one in following condition or any group
Close:
Condition 1: if it is determined that user behaviors log in, the daily record quantity of certain user be more than or equal to predetermined number
Threshold value, is greater than or equal to 5, then the daily record of this user is as training sample.
Condition 2: if it is determined that user behaviors log in, certain daily record comprised indication a certain consumption be intended to pass
Keyword accounting is more than or equal to preset ratio threshold value, then using this daily record as training sample.
Condition 3: if it is determined that user behaviors log in, certain daily record occurring, the occurrence number of most brands accounts for
In this daily record, the ratio of the occurrence number of all brands exceedes preset ratio threshold value, then using this daily record as instruction
Practice sample.Certain daily record such as occurs in that the brands such as " association ", " Samsung ", " Fructus Mali pumilae ", but " Herba Marsileae Quadrifoliae
It is really " that most brand occurs, if the occurrence number of " Fructus Mali pumilae " accounts for going out of all brands in this daily record
Occurrence number ratio more than 0.5, then illustrates that the intention of user compares concentration, using this daily record as training sample,
Otherwise explanation user view ratio is relatively decentralized, it is likely that is exactly at will to browse, just abandons this daily record.
Step 104: utilize training sample train classification models, obtain can recognize that purchase move ahead into
The consumption intention assessment model that the setting consumer field of behavior after purchase is corresponding.
The present invention is not limiting as the type of disaggregated model, arbitrary two classification in prior art can be used
Model, such as SVM(support vector machine) model etc., the consumption intention assessment model energy finally trained
Enough it is identified according to the probability of each type.
The feature for user used when utilizing training sample train classification models can at least include
One in following characteristics or combination in any:
Daily record quantity；
Daily record quantity shared by the brand that occurrence number is most；
The daily record quantity shared by brand that occurrence number time is many；
Daily record quantity shared by the model that occurrence number is most；
The daily record quantity shared by model that occurrence number time is many；
The key word accounting that indication consumption is intended to；
The appearance position of the key word that indication consumption is intended to；
There is the daily record accounting of query；
The key word accounting of the indication user view comprised in query；
Duration is crossed in daily record；
The maximum of daily record shared by same brand crosses over duration；
The maximum of daily record shared by same model crosses over duration.
Except one group of training sample training classification mould of the employing shown in above-mentioned steps 103 and step 104
Type, thus obtain outside the method for consumption intention assessment model, in order to improve the recognition accuracy of model,
One is also provided for preferred embodiment at this:
In step 103, select to meet N group training number respectively from the user behaviors log that step 102 determines
N number of user behaviors log set according to screening conditions, say, that pre-set N group training data screening bar
Part, N is the positive integer more than 1, and each group of training data screening conditions can obtain a user behaviors log
Set, N group just obtains N number of user behaviors log set.It is right that these N group training data screening conditions can use
The modes such as above-mentioned 3 condition setting varying number threshold values, proportion threshold value obtain, such as, divide in condition 2
Other Set scale threshold value is: 0,0.1,0.2,0.3,0.4,0.5,0.6 etc., sets respectively in condition 3
Putting proportion threshold value is 0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9.
Using N number of user behaviors log set as training sample train classification models, obtain N number of candidate's mould
Type；Then from this N number of candidate family, a conduct consumption intention assessment model of optimum is selected, at this
Optimum one can be candidate family meets preset recall rate require and accuracy rate is the highest one.Tool
Body ground, can be utilized respectively N number of candidate family and classify test sample, according to test sample
Classification results determines recall rate and the accuracy rate of each candidate family.
Further, since arranging of probability threshold value can directly affect the classification results distributing model in disaggregated model
(so-called probability threshold value refers to when disaggregated model prediction object A belongs to the probability of classification results X more than being somebody's turn to do
During probability threshold value, it is believed that this object A belongs to classification results X), therefore, obtaining N number of candidate family
After, by carrying out the setting of different probability threshold values respectively for N number of candidate family, can obtain having not
Candidate family with probability threshold value, it is assumed that respectively N number of candidate family is respectively provided with probability threshold value be 0.5,
0.6, the setting of 0.7,0.8,0.9 these five kinds of probability threshold values, then actually obtained 5N time
Modeling type, then select satisfied default recall rate requirement and accuracy rate the highest from this 5N candidate family
One as consumption intention assessment model.
Embodiment two,
Fig. 3 identifies, for what the embodiment of the present invention two provided, the method flow diagram that consumption is intended to, and the method is base
Consume what intention assessment model realized in what mode shown in embodiment one was set up, as it is shown on figure 3, the party
Method comprises the following steps:
Step 301: determine the consumer field of user to be identified.
This step can in different ways according to different application scenarios, can be according to the current institute of user
It is determined at type of webpage, such as the digital product column during active user checks certain website, that
Just can directly determine that the consumer field of user to be identified is for digital consumer product area；Or, Ke Yigen
According to the consumer field belonging to the user behaviors log identification behavior daily record of user to be identified of nearly a period of time, determine
Consumer field belonging to user behaviors log can determine mode to use theme of the prior art, the most superfluous at this
State.The duration of usual described nearly a period of time arranges and sets up the screening that consumption intention assessment model uses
When the duration of corelation behaviour daily record arranges consistent, it is possible to there is higher accuracy rate；Or, can basis
The content that user is currently entered determines, such as when user uses search engine input Samsung galaxy S4,
Content according to user's input is it is determined that go out consumer field for digital consumer product area.
Step 302: the consumption intention assessment model pair that the consumer field that utilizes step 301 to identify is corresponding
The user behaviors log of user to be identified is classified, and obtains the consumption of user to be identified and is intended that before purchase still
After purchase.
In this step, need to extract the feature of the user behaviors log of user to be identified, consume intention assessment mould
Type actually utilizes the feature of the user behaviors log of user to be identified to carry out classifying, the spy herein extracted
Levying should be consistent with the feature used when train classification models in embodiment one step 104, and which specifically can use
A little features do not repeat them here.The classification foundation of consumption intention assessment model is actually from use to be identified
The feature that the user behaviors log at family extracts, thus the consumption obtaining user to be identified is intended that before purchase still
After purchase.
After the consumption identifying user is intended to, it becomes possible to as disclosed at step 303, according to the consumption of user
It is intended to corresponding exhibition strategy targetedly to user's exhibition information.Specifically can apply to but be not limited to
Following two scene:
The first scene: be intended to according to the consumption of user to be identified, show with to be identified to user to be identified
The consumption of user is intended to the promotion message of corresponding types.If before the consumption of such as user is intended that purchase,
Then can buy information, rate of exchange information, contrast evaluation and test information etc. to user's Recommendations, if user
After consumption is intended that purchase, then can to user recommend gains in depth of comprehension information, relevant peripheral merchandise news,
The information such as information after sale.
The second scene: when user to be identified uses search engine, can be according to customer consumption to be identified
It is intended to scan for the sequence of result, the consumption of user to be identified is intended to corresponding information at Search Results
In sequence in advance.If before the consumption such as identifying user is intended that purchase, then can be by search knot
In Guo about goods purchase information, rate of exchange information, contrast evaluation and test information webpage sequence in advance, if
Identify after the consumption of user is intended that purchase, then can by Search Results about use gains in depth of comprehension information,
The sequence of the webpages such as relevant peripheral merchandise news, after sale information is in advance.
It is above the detailed description that method provided by the present invention is carried out, below by embodiment three to this
What invention provided identifies that the device that consumption is intended to is described in detail.
Embodiment three,
Fig. 4 identifies, for what the embodiment of the present invention three provided, the structure drawing of device that consumption is intended to, and this device generally sets
Putting at server end, as shown in Figure 4, this device includes modeling unit 00 and recognition unit 10.
Wherein, modeling unit 00 includes: the first screening subelement 01, pattern analysis subelement 02, second
Screening subelement 03 and model training subelement 04.
First the first screening subelement 01 filters out from the historical behavior daily record of each user and sets consumer field
Corelation behaviour daily record.The mode carrying out screening at this can use but the one that is not limited in the following manner or group
Close: mode based on Keywords matching, and mode based on template matching.
Mode based on Keywords matching is, excavates in advance and sets the lists of keywords that consumer field is corresponding,
The lists of keywords that each user behaviors log is corresponding with setting consumer field respectively matches, and filters out and comprises key
The user behaviors log of the key word in word list.Here lists of keywords can be one or more, can screen
Go out to comprise the user behaviors log of key word in any one lists of keywords, or filter out and comprise several simultaneously
The user behaviors log of the key word in lists of keywords.
Mode based on template matching is, excavates in advance and sets the expression template that consumer field is corresponding, will be each
User behaviors log mates with this expression template respectively, filters out the user behaviors log mated with this expression template.
Then, pattern analysis subelement 02 is based on the corelation behaviour day filtering out the first screening subelement 01
The BMAT that will is carried out, determine corresponding buy move ahead into user behaviors log and corresponding buy after behavior
User behaviors log.
Pattern analysis subelement 02 can be based on the behavior manually carried out the corelation behaviour daily record filtered out herein
Pattern analysis, it is also possible to automatically analyzed by pattern analysis subelement 02, i.e. according in corelation behaviour daily record
The key word that the indication consumption comprised is intended to carries out the analysis of behavioral pattern, the key word bag that indication consumption is intended to
The key word of intention after including the key word being intended to before indicating purchase or indicating purchase.
Afterwards, the second screening subelement 03 selects full from the user behaviors log that pattern analysis subelement 02 determines
The user behaviors log of foot training data screening conditions is as training sample.Training data screening conditions therein include
One in following condition or combination in any:
Condition 1: the daily record quantity that the user behaviors log of user comprises is more than or equal to predetermined number threshold value.
Condition 2: the key word accounting that the indication consumption that user behaviors log is comprised is intended to is more than or equal to preset ratio
Threshold value, indicates that the key word that consumption is intended to is to indicate intention after the front key word being intended to of purchase or indication purchase
Key word.
Condition 3: occur in user behaviors log that the occurrence number of most brands accounts for all brands in behavior daily record
The ratio of occurrence number exceedes preset ratio threshold value.
It addition, the key word being intended to realize above-mentioned indication consumption excavates, this modeling unit 00 also includes meaning
Graph key word excavates subelement 05, for excavating indication consumption meaning according to the mode shown in following operation A1-A4
The key word of figure:
Operate A1, for setting the seed words that consumer field determines that indication consumption is intended to.Seed words is permissible herein
The most manually arrange, it is common that arrange for a certain concrete consumer field, but a consumer field would generally be wrapped
Containing multiple subclasses, when selected seed word, preferably it is chosen in each subclass meaning before the indication being all suitable for is bought
The seed words being intended to after the seed words of figure and purchase.Such as, for digital consumer product area, electricity is generally included
Each subclasses such as brain, camera, mobile phone, before this can be by indication purchase in the subclasses such as computer, camera and mobile phone
The key word being intended to takes to occur simultaneously and is used as the seed words that in digital consumer product area, indication is intended to before buying.
Operate A2, utilize seed words that the sample daily record setting consumer field is classified, obtain buying front sample
Sample daily record after this daily record and purchase.
Specifically, choose some sample daily records setting consumer field in advance, if certain sample daily record is wrapped
The seed words that before buying containing indication, consumption is intended to is more than the shape comprising the seed words consuming intention after indication is bought
Condition meets pre-conditioned, then it is assumed that this sample daily record is sample daily record before buying；If in certain sample daily record
The unnecessary seed words comprising indication purchase front consumption intention of seed words of intention is consumed after comprising indication purchase
Situation meets pre-conditioned, then it is assumed that this sample daily record is sample daily record after buying.Here pre-conditioned
Can use: difference is more than preset difference value threshold value, difference accounts for the seed words sum comprised in this sample daily record
The ratio of amount exceedes preset ratio threshold value, etc..
Operate A3, respectively sample daily record after sample daily record before buying and purchase carried out the mistake of participle and stop word
After filter, add up the word frequency of each word.
The stop word wherein filtered can include but not limited to following at least one: with brand, model
Relevant word, the word insignificant to semantic meaning representation such as function word, auxiliary word, modal particle, there are letter, numeral
Or the noise word such as a Chinese character.
Operation A4, based on each word respectively sample daily record and the word frequency in sample daily record after buying the most before purchase,
Determine the key word that the key word that indication is intended to before buying is intended to after buying with indication.
If the word frequency in the sample daily record before purchase of certain word is the highest, but after point of purchase in sample daily record
Word frequency is the lowest or be 0, it is determined that this word is the key word being intended to before indication is bought, equally, if
Word frequency in the sample daily record after point of purchase of certain word is the highest, but the word frequency in sample daily record is the lowest before purchase
Or it is 0, it is determined that this word is the key word being intended to after indication is bought.
Finally, model training subelement 04 extracts features training disaggregated model from training sample, and obtaining can
Identify buy move ahead into buy after consumption intention assessment model corresponding to the setting consumer field of behavior.Its
The disaggregated model used can be arbitrary two disaggregated models, such as SVM model in prior art, finally instructs
The consumption intention assessment model practised can be identified according to the probability of each type.
Model training subelement 04 when train classification models from training sample extract feature at least include but
The one being not limited in following characteristics or combination in any:
Daily record quantity；
Daily record quantity shared by the brand that occurrence number is most；
The daily record quantity shared by brand that occurrence number time is many；
Daily record quantity shared by the model that occurrence number is most；
The daily record quantity shared by model that occurrence number time is many；
The key word accounting that indication consumption is intended to；
The appearance position of the key word that indication consumption is intended to；
There is the daily record accounting of query；
The key word accounting of the indication user view comprised in query；
Duration is crossed in daily record；
The maximum of daily record shared by same brand crosses over duration；
The maximum of daily record shared by same model crosses over duration.
Except the second above-mentioned screening subelement 03 and model training subelement 04 use one group of training sample instruction
Practice disaggregated model, thus obtain outside the method for consumption intention assessment model, accurate in order to improve the identification of model
Really rate, also provides for one preferred embodiment at this:
Second screening subelement 03 selects to meet respectively N from the user behaviors log that pattern analysis subelement 02 determines
N number of user behaviors log set of group training data screening conditions.Above-mentioned N group training data screening conditions can be adopted
Obtain, such as in condition 2 with to modes such as above-mentioned 3 condition setting varying number threshold values, proportion threshold value
In be respectively provided with proportion threshold value and be: 0,0.1,0.2,0.3,0.4,0.5,0.6 etc., in condition 3 point
Other Set scale threshold value is 0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9.
Each user behaviors log set is trained by model training subelement 04 as one group of training sample
Disaggregated model, obtains N number of candidate family, selects one of optimum as setting from N number of candidate family
The consumption intention assessment model that consumer field is corresponding.Specifically, N number of candidate family pair can be utilized respectively
Test sample is classified, according to the classification results of test sample is determined each candidate family recall rate and
Accuracy rate.
After consumption intention assessment model is set up, it is possible to for the customer consumption of recognition unit 10
Intention assessment.As shown in Figure 4, recognition unit 10 includes: field determines subelement 11 and category of model
Subelement 12, it is also possible to the information that farther includes shows subelement 13.
First field determines that subelement 11 determines the consumer field of user to be identified.According to different application scenarios,
Field determines that subelement 11 can be in different ways: can be currently located type of webpage according to user and carry out
Determine, such as the digital product column during active user checks certain website, then just can directly determine
The consumer field of user to be identified is digital consumer product area；Or, can be to be identified according to nearly a period of time
The consumer field belonging to user behaviors log identification behavior daily record of user, determines the consumption neck belonging to user behaviors log
Territory can determine mode to use theme of the prior art, does not repeats them here.Usual described nearly a period of time
Duration arrange and set up the duration of screening corelation behaviour daily record that consumption intention assessment model uses and arrange one
During cause, it is possible to there is higher accuracy rate；Or, can determine according to the content that user is currently entered, example
As when user use search engine input Samsung galaxy S4 time, according to user input content it is determined that
Go out consumer field for digital consumer product area.
Then the consumer field that category of model subelement 12 utilizes field to determine that subelement 11 is determined is corresponding
Consumption intention assessment model, to user to be identified in nearly a period of time in the associated row of the consumer field determined
Classify for daily record, after obtaining before the consumption of user to be identified is intended that purchase or buying.Category of model
Unit needs to extract the feature of the user behaviors log of user to be identified, and consumption intention assessment model actually utilizes
The feature of the user behaviors log of user to be identified carries out classifying.Its feature extracted should be with model training list
The feature that unit 04 extracts is consistent.
Further, above-mentioned information shows that subelement 13 just can be right according to the consumption intention of user to be identified
The exhibition strategy answered is to user's exhibition information to be identified.Specifically, can apply to but be not limited to following two
Scene:
The first scene: be intended to according to the consumption of user to be identified, show and use to be identified to user to be identified
The consumption at family is intended to the promotion message of corresponding types.If before the consumption of such as user is intended that purchase, then may be used
To buy information, rate of exchange information, contrast evaluation and test information etc. to user's Recommendations, if the consumption meaning of user
After figure is purchase, then can recommend gains in depth of comprehension information, relevant peripheral merchandise news, after sale information to user
Etc. information.
The second scene: in the Search Results that user to be identified shows, the consumption of user to be identified is anticipated
The sequence in Search Results of the information of figure correspondence is in advance.If the consumption such as identifying user is intended that purchase
Before buying, then can by Search Results about goods purchase information, rate of exchange information, contrast evaluation and test information net
The sequence of page in advance, if after the consumption identifying user is intended that purchase, then can will be closed in Search Results
In using the sequence of the webpage such as gains in depth of comprehension information, relevant peripheral merchandise news, after sale information in advance.
The present invention provide said method and device can be widely applicable for Webpage search, commercial articles searching,
The fields such as advertisement promotion, show to user and more meet the content that customer consumption is intended to, and improve accuracy, no
It is only capable of and better meets user's request, reduce the invalid content interference to user, can also improve simultaneously
Effective commercial activity conversion ratio.
In several embodiments provided by the present invention, it should be understood that disclosed apparatus and method,
Can realize by another way.Such as, device embodiment described above is only schematically,
Such as, the division of described unit, it is only a kind of logic function and divides, actual can have additionally when realizing
Dividing mode.
It addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit
In, it is also possible to it is that unit is individually physically present, it is also possible to two or more unit are integrated in one
In individual unit.Above-mentioned integrated unit both can realize to use the form of hardware, it would however also be possible to employ hardware adds
The form of SFU software functional unit realizes.
The above-mentioned integrated unit realized with the form of SFU software functional unit, can be stored in a computer-readable
Take in storage medium.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions in order to
Make a computer equipment (can be personal computer, server, or the network equipment etc.) or process
Device (processor) performs the part steps of method described in each embodiment of the present invention.And aforesaid storage medium
Including: USB flash disk, portable hard drive, read only memory (Read-Only Memory, ROM), random access memory are deposited
Reservoir (Random Access Memory, RAM), magnetic disc or CD etc. are various can store program code
Medium.
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all at this
Within bright spirit and principle, any modification, equivalent substitution and improvement etc. done, should be included in this
Within the scope of bright protection.
Claims (18)
1. the method identifying that consumption is intended to, it is characterised in that the method includes:
The process of foundation consumption intention assessment model:
S11, filter out from the historical behavior daily record of each user set consumer field corelation behaviour daily record；
S12, based on the BMAT that the corelation behaviour daily record filtered out is carried out, before determining corresponding purchase
The user behaviors log of behavior after the user behaviors log of behavior and corresponding purchase；
S13, from the user behaviors log that step S12 determines, select to meet the user behaviors log of training data screening conditions
As training sample；
S14, from training sample, extract features training disaggregated model, obtain can recognize that purchase move ahead into
The consumption intention assessment model that after purchase, the described setting consumer field of behavior is corresponding；
The process that identification consumption is intended to:
S21, determine the consumer field of user to be identified；
S22, utilize consumption intention assessment model corresponding to the consumer field determined, to institute in nearly a period of time
State user to be identified to classify in the corelation behaviour daily record of the described consumer field determined, obtain to be identified
Before the consumption of user is intended that purchase or after purchase.
Method the most according to claim 1, it is characterised in that described step S11 specifically includes:
N the lists of keywords corresponding with described setting consumer field respectively by the historical behavior daily record of each user
Matching, filter out and comprise the user behaviors log of key word in described n lists of keywords simultaneously, described n is
Positive integer；Or,
The expression template that the historical behavior daily record of each user is corresponding with described setting consumer field respectively is matched,
Filter out the user behaviors log mated with described expression template.
Method the most according to claim 1, it is characterised in that to filtering out in described step S12
The BMAT that corelation behaviour daily record is carried out is:
Manually the corelation behaviour daily record filtered out is carried out BMAT；Or,
The key word being intended to according to the indication consumption comprised in corelation behaviour daily record carries out the analysis of behavioral pattern, institute
State indication consumption be intended to key word include indication buy before be intended to key word or indication buy after be intended to pass
Keyword.
Method the most according to claim 1, it is characterised in that described training data screening conditions include
One in following condition or combination in any:
The daily record quantity that the user behaviors log of user comprises is more than or equal to predetermined number threshold value；
The key word accounting that the indication consumption that user behaviors log is comprised is intended to is more than or equal to preset ratio threshold value, institute
State indication consumption be intended to key word be indication buy before be intended to key word or indication buy after be intended to key
Word；
User behaviors log occurring, what the occurrence number of most brands accounted for all brands in behavior daily record goes out occurrence
The ratio of number exceedes preset ratio threshold value.
5. according to the method described in claim 3 or 4, it is characterised in that the pass that described indication consumption is intended to
Keyword excavates in the following way:
A1, for described setting consumer field determine indication consumption be intended to seed words；
A2, utilize described seed words that the sample daily record of described setting consumer field is classified, bought
Sample daily record after front sample daily record and purchase；
A3, respectively to sample daily record before buying with after sample daily record carries out the filtration of participle and stop word after buying,
Add up the word frequency of each word；
A4, based on each word respectively sample daily record and the word frequency in sample daily record after buying the most before purchase, determine
The key word that the key word that indication is intended to before buying is intended to after buying with indication.
Method the most according to claim 1, it is characterised in that from training sample in described step S14
The feature extracted in Ben at least includes but not limited to the one in following characteristics or combination in any:
Daily record quantity；
Daily record quantity shared by the brand that occurrence number is most；
The daily record quantity shared by brand that occurrence number time is many；
Daily record quantity shared by the model that occurrence number is most；
The daily record quantity shared by model that occurrence number time is many；
The key word accounting that indication consumption is intended to；
The appearance position of the key word that indication consumption is intended to；
There is the daily record accounting of query；
The key word accounting of the indication user view comprised in query；
Duration is crossed in daily record；
The maximum of daily record shared by same brand crosses over duration；
The maximum of daily record shared by same model crosses over duration.
7. according to the method described in claim 1,4 or 6, it is characterised in that described step S13 is concrete
Including: from the user behaviors log that step S12 determines, select to meet respectively the N of N group training data screening conditions
Individual user behaviors log set；
Described step S14 specifically includes: each user behaviors log set instructed as one group of training sample
Practice disaggregated model, obtain N number of candidate family, select to meet from described N number of candidate family and preset recall rate
Require and accuracy rate the highest is as consumption intention assessment model corresponding to described setting consumer field.
Method the most according to claim 1, it is characterised in that after described step S22, also wrap
Include:
S23, it is intended to corresponding exhibition strategy to described user's exhibition to be identified according to the consumption of described user to be identified
Show information.
Method the most according to claim 8, it is characterised in that described step S23 specifically includes:
It is intended to according to the consumption of described user to be identified, shows and described use to be identified to described user to be identified
The consumption at family is intended to the promotion message of corresponding types；Or,
In the Search Results that described user to be identified shows, it is right the consumption of described user to be identified to be intended to
The information answered sequence in Search Results is in advance.
10. the device identifying that consumption is intended to, it is characterised in that this device includes modeling unit and identification
Unit；
Described modeling unit includes:
First screening subelement, sets consumer field for filtering out from the historical behavior daily record of each user
Corelation behaviour daily record；
Pattern analysis subelement, is used for based on the BMAT carrying out the corelation behaviour daily record filtered out,
Determine corresponding buy move ahead into user behaviors log and corresponding buy after the user behaviors log of behavior；
Second screening subelement, satisfied for selecting from the user behaviors log that described pattern analysis subelement determines
The user behaviors log of training data screening conditions is as training sample；
Model training subelement, for extracting features training disaggregated model from training sample, obtains knowing
Do not go out to buy move ahead into buy after consumption intention assessment model corresponding to the described setting consumer field of behavior；
Described recognition unit includes:
Field determines subelement, for determining the consumer field of user to be identified；
Category of model subelement is corresponding for the consumer field utilizing described field to determine that subelement is determined
Consumption intention assessment model, to user to be identified described in nearly a period of time in the described consumer field determined
Corelation behaviour daily record classify, after obtaining before the consumption of user to be identified is intended that purchase or buying.
11. devices according to claim 10, it is characterised in that described first screening subelement is concrete
Perform:
N the lists of keywords corresponding with described setting consumer field respectively by the historical behavior daily record of each user
Matching, filter out and comprise the user behaviors log of key word in described n lists of keywords simultaneously, described n is
Positive integer；Or,
By expression template phase corresponding with described setting consumer field respectively for the historical behavior daily record of each user
Join, filter out the user behaviors log mated with described expression template.
12. devices according to claim 10, it is characterised in that described pattern analysis subelement based on
The artificial BMAT that the corelation behaviour daily record filtered out is carried out, or, according to corelation behaviour daily record
In the key word that is intended to of the indication consumption that comprises carry out the analysis of behavioral pattern, the pass that described indication consumption is intended to
Keyword include indication buy before be intended to key word or indication buy after be intended to key word.
13. devices according to claim 10, it is characterised in that described training data screening conditions bag
Include the one in following condition or combination in any:
The daily record quantity that the user behaviors log of user comprises is more than or equal to predetermined number threshold value；
The key word accounting that the indication consumption that user behaviors log is comprised is intended to is more than or equal to preset ratio threshold value, institute
State indication consumption be intended to key word be indication buy before be intended to key word or indication buy after be intended to key
Word；
User behaviors log occurring, what the occurrence number of most brands accounted for all brands in behavior daily record goes out occurrence
The ratio of number exceedes preset ratio threshold value.
14. according to the device described in claim 12 or 13, it is characterised in that described modeling unit is also wrapped
Include intention key word and excavate subelement, for excavating the key word that described indication consumption is intended in such a way:
A1, for described setting consumer field determine indication consumption be intended to seed words；
A2, utilize described seed words that the sample daily record of described setting consumer field is classified, bought
Sample daily record after front sample daily record and purchase；
A3, respectively to sample daily record before buying with after sample daily record carries out the filtration of participle and stop word after buying,
Add up the word frequency of each word；
A4, based on each word respectively sample daily record and the word frequency in sample daily record after buying the most before purchase, determine
The key word that the key word that indication is intended to before buying is intended to after buying with indication.
15. devices according to claim 10, it is characterised in that described model training subelement is in instruction
The feature and the described category of model subelement that extract from training sample when practicing disaggregated model are utilizing consumption meaning
Figure identify model when classifying from described user to be identified in the corelation behaviour of the described consumer field determined
The feature extracted in daily record at least includes but not limited to the one in following characteristics or combination in any:
Daily record quantity；
Daily record quantity shared by the brand that occurrence number is most；
The daily record quantity shared by brand that occurrence number time is many；
Daily record quantity shared by the model that occurrence number is most；
The daily record quantity shared by model that occurrence number time is many；
The key word accounting that indication consumption is intended to；
The appearance position of the key word that indication consumption is intended to；
There is the daily record accounting of query；
The key word accounting of the indication user view comprised in query；
Duration is crossed in daily record；
The maximum of daily record shared by same brand crosses over duration；
The maximum of daily record shared by same model crosses over duration.
16. according to the device described in claim 10,13 or 15, it is characterised in that described second sieve
Subelement is selected to select to meet N group training number respectively from the user behaviors log that described pattern analysis subelement determines
N number of user behaviors log set according to screening conditions；
Described model training subelement is using each user behaviors log set as one group of training sample training point
Class model, obtains N number of candidate family, selects to meet and preset recall rate requirement from described N number of candidate family
And accuracy rate the highest is as consumption intention assessment model corresponding to described setting consumer field.
17. devices according to claim 10, it is characterised in that described recognition unit also includes: letter
Breath shows subelement, treats to described for being intended to corresponding exhibition strategy according to the consumption of described user to be identified
Identify user's exhibition information.
18. devices according to claim 17, it is characterised in that described information shows that subelement is concrete
Perform:
It is intended to according to the consumption of described user to be identified, shows and described use to be identified to described user to be identified
The consumption at family is intended to the promotion message of corresponding types；Or,
In the Search Results that described user to be identified shows, it is right the consumption of described user to be identified to be intended to
The information answered sequence in Search Results is in advance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310301375.7A CN103412882B (en) | 2013-07-18 | 2013-07-18 | A kind of method and device identifying that consumption is intended to |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310301375.7A CN103412882B (en) | 2013-07-18 | 2013-07-18 | A kind of method and device identifying that consumption is intended to |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103412882A CN103412882A (en) | 2013-11-27 |
| CN103412882B true CN103412882B (en) | 2016-11-23 |
Family
ID=49605894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310301375.7A Active CN103412882B (en) | 2013-07-18 | 2013-07-18 | A kind of method and device identifying that consumption is intended to |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103412882B (en) |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104778176A (en) * | 2014-01-13 | 2015-07-15 | 阿里巴巴集团控股有限公司 | Data search processing method and device |
| CN104933075A (en) * | 2014-03-20 | 2015-09-23 | 百度在线网络技术（北京）有限公司 | User attribute predicting platform and method |
| EP3123356A4 (en) * | 2014-03-26 | 2017-09-06 | Microsoft Technology Licensing, LLC | Client intent in integrated search environment |
| CN105094305B (en) * | 2014-05-22 | 2018-05-18 | 华为技术有限公司 | Identify method, user equipment and the Activity recognition server of user behavior |
| CN105335391B (en) * | 2014-07-09 | 2019-02-15 | 阿里巴巴集团控股有限公司 | The treating method and apparatus of searching request based on search engine |
| CN104462024B (en) * | 2014-10-29 | 2018-07-13 | 百度在线网络技术（北京）有限公司 | The method and apparatus for generating dialogue action policy model |
| CN104835057A (en) * | 2015-04-02 | 2015-08-12 | 百度在线网络技术（北京）有限公司 | Method and device for obtaining consumption feature information of network user |
| CN104951433B (en) * | 2015-06-24 | 2018-01-23 | 北京京东尚科信息技术有限公司 | The method and system of intention assessment is carried out based on context |
| CN106649312B (en) * | 2015-10-29 | 2019-10-29 | 北京北方华创微电子装备有限公司 | The analysis method and system of journal file |
| CN105718564A (en) * | 2016-01-20 | 2016-06-29 | 清华大学 | Promotion behavior detection method and apparatus |
| CN107133259A (en) * | 2017-03-22 | 2017-09-05 | 北京晓数聚传媒科技有限公司 | A kind of searching method and device |
| CN107798622B (en) * | 2017-10-18 | 2021-06-29 | 北京京东尚科信息技术有限公司 | Method and device for identifying user intention |
| CN109685537B (en) * | 2017-10-18 | 2021-02-26 | 北京京东尚科信息技术有限公司 | User behavior analysis method, device, medium and electronic equipment |
| CN109167816B (en) * | 2018-08-03 | 2021-11-16 | 广州虎牙信息科技有限公司 | Information pushing method, device, equipment and storage medium |
| CN109190044B (en) * | 2018-09-10 | 2020-11-20 | 北京百度网讯科技有限公司 | Personalized recommendation method, device, server and medium |
| CN109492104B (en) * | 2018-11-09 | 2023-01-31 | 北京汇钧科技有限公司 | Training method, classification method, system, device and medium of intention classification model |
| CN109558544B (en) * | 2018-12-12 | 2021-04-27 | 拉扎斯网络科技（上海）有限公司 | Sorting method and device, server and storage medium |
| CN109726764A (en) * | 2018-12-29 | 2019-05-07 | 北京航天数据股份有限公司 | A kind of model selection method, device, equipment and medium |
| CN111400495A (en) * | 2020-03-17 | 2020-07-10 | 重庆邮电大学 | Video bullet screen consumption intention identification method based on template characteristics |
| CN111738017A (en) * | 2020-06-24 | 2020-10-02 | 深圳前海微众银行股份有限公司 | Intention identification method, device, equipment and storage medium |
| CN113190599B (en) * | 2021-06-30 | 2021-09-28 | 平安科技（深圳）有限公司 | Processing method, device and equipment for application user behavior data and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908184A (en) * | 2009-06-04 | 2010-12-08 | 维鹏信息技术(上海)有限公司 | Control method and system for distributing information through multiple associated terminals |
| JP2012094131A (en) * | 2010-09-30 | 2012-05-17 | Rakuten Inc | Electronic commerce server, electronic commerce method, electronic commerce program and computer-readable recording medium for recording program |
| CN103034680A (en) * | 2012-11-19 | 2013-04-10 | 北京奇虎科技有限公司 | Data interaction method and device for terminal device |
| CN103208073A (en) * | 2012-01-17 | 2013-07-17 | 阿里巴巴集团控股有限公司 | Method and device for obtaining recommend commodity information and providing commodity information |
-
2013
- 2013-07-18 CN CN201310301375.7A patent/CN103412882B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101908184A (en) * | 2009-06-04 | 2010-12-08 | 维鹏信息技术(上海)有限公司 | Control method and system for distributing information through multiple associated terminals |
| JP2012094131A (en) * | 2010-09-30 | 2012-05-17 | Rakuten Inc | Electronic commerce server, electronic commerce method, electronic commerce program and computer-readable recording medium for recording program |
| CN103208073A (en) * | 2012-01-17 | 2013-07-17 | 阿里巴巴集团控股有限公司 | Method and device for obtaining recommend commodity information and providing commodity information |
| CN103034680A (en) * | 2012-11-19 | 2013-04-10 | 北京奇虎科技有限公司 | Data interaction method and device for terminal device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103412882A (en) | 2013-11-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103412882B (en) | A kind of method and device identifying that consumption is intended to | |
| JP6582085B2 (en) | Method and apparatus for generating web page content | |
| CN102279851B (en) | Intelligent navigation method, device and system | |
| CN103218719B (en) | A kind of e-commerce website air navigation aid and system | |
| CN105045901B (en) | The method for pushing and device of search key | |
| KR100786795B1 (en) | Internet advertising service system and method thereof | |
| CN105447186B (en) | A kind of user behavior analysis system based on big data platform | |
| CN105159930B (en) | The method for pushing and device of search key | |
| CN109165301A (en) | Video cover selection method, device and computer readable storage medium | |
| CN107909392B (en) | Stimulation behavior sensitivity prediction method, device, equipment and readable storage medium | |
| CN108319376B (en) | Input association recommendation method and device for optimizing commercial word promotion | |
| CN106445963B (en) | Advertisement index keyword automatic generation method and device of APP platform | |
| CN106846061A (en) | Potential user's method for digging and device | |
| CN104835057A (en) | Method and device for obtaining consumption feature information of network user | |
| CN110246007A (en) | A kind of Method of Commodity Recommendation and device | |
| CN107330459A (en) | A kind of data processing method, device and electronic equipment | |
| CN110020128A (en) | A kind of search result ordering method and device | |
| CN106326318B (en) | Searching method and device | |
| CN102279872A (en) | Inquiring intention identification drived by search results | |
| CN109685540A (en) | Electric business platform and its commercial promotions control method, equipment, readable storage medium storing program for executing | |
| CN103778553A (en) | Commodity attribute recommendation method and commodity attribute recommendation system | |
| CN106919575A (en) | application program searching method and device | |
| CN103377249A (en) | Keyword putting method and system | |
| CN102811207A (en) | Network information pushing method and system | |
| CN105740415A (en) | Label position weight and self-learning based tendering and bidding good friend recommendation system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |