The content of the invention
In view of this, the present invention provide it is a kind of for electric business full deactivation it is dynamic gather single commercial product recommending
Method, can make commercial product recommending technology in combination with commercial promotions activity, make consumer obtain it
It is satisfied to gather single Recommendations, the validity of electric business platform advertising campaign is improved, increase businessman and be full of
The experience effect of user is improved while sharp.
For achieving the above object, the invention provides it is a kind of for the full deactivation of electric business it is dynamic gather single business
The method that product are recommended.
The gather method of single commercial product recommending dynamic for the full deactivation of electric business of the present invention includes:By poly-
Commodity sales promotion is divided into different equivalence classes by the mode of class;Different equivalence classes are divided into similar
Equivalence class and related equivalence class, and calculate the demand recommendation grade of each related equivalence class;According to not
With Generalization bounds, select from similar equivalence class and/or related equivalence class to consumer recommend with
Fill and completely subtract the commodity sales promotion or grouping of commodities of difference, and export recommendation results collection, wherein, will
Different equivalence classes divide into similar equivalence class and related equivalence class to be included:If consumer has selected business
Product, then it is similar equivalence class that consumer has selected the equivalence class belonging to commodity, and remaining equivalence class is
Related equivalence class;If the non-selected commodity of consumer, consumer is in the recent period belonging to the commodity of purchase
Equivalence class is similar equivalence class, and remaining equivalence class is related equivalence class.
Alternatively, commodity sales promotion is divided into different equivalence classes by way of cluster includes:
Semantic Clustering is carried out according to the attribute of commodity sales promotion, commodity sales promotion is divided into not homology equivalence
Class.
Alternatively, calculating the demand recommendation grade of each related equivalence class includes:If consumer selects
Commodity are selected, then the relevance grades of commodity and each related equivalence class has been selected according to consumer and has been disappeared
Two factors of inclined happiness degree of the person of expense to each related equivalence class, are counted using the weighting of commercial product recommending technology
Calculate the demand recommendation grade of each related equivalence class;If the non-selected commodity of consumer, according to consumption
The recent commodity of purchase of person are to the relevance grades of each related equivalence class and consumer to each correlation etc.
Two factors of inclined happiness degree of valency class, using each related equivalence class of commercial product recommending technology weighted calculation
Demand recommendation grade.
Alternatively, according to different Generalization bounds, from similar equivalence class and/or related equivalence class
Select to recommend to include under use to fill the commodity sales promotion or grouping of commodities that completely subtract difference to consumer
One or more in row strategy:Strategy one:From the high related equivalence class of demand recommendation grade
Recommended price is close to the commodity or grouping of commodities for completely subtracting difference, completely subtract difference to fill;Strategy two:
Similar commodity are found in the similar equivalence class belonging to a certain commodity that consumer has selected and substitutes institute
The a certain commodity for having selected are stated, completely subtract difference to fill;Strategy three:Utilization strategies two select phase
Substituted like commodity, when select the close consumer of similar commodity inclined happiness degree but price not
Can fill when completely subtracting difference, Utilization strategies one are pushed away from the high related equivalence class of demand recommendation grade
Commodity or grouping of commodities that close prices completely subtracts difference are recommended, completely subtract difference to fill.
Alternatively, described tactful one also includes:Selection demand recommendation grade highest is related of equal value
A commodity in class are added in the currentElement of recommendation results collection, and judge the currentElement
In commodity price sum whether less than difference is completely subtracted, if less than difference is completely subtracted, continuing to select
A commodity add the currentElement in demand recommendation grade highest correlation equivalence class, repeat
Perform current step;If more than difference is completely subtracted, judging the commodity valency in the currentElement
Whether lattice sum is more than recommendation threshold value, if being more than recommendation threshold value, in deleting the currentElement
A commodity being eventually adding, and repeat current step;If less than recommendation threshold value,
Then the currentElement commodity are added and terminated;Repeat the above steps, until obtaining containing multiple units
The recommendation results collection of element.
Technology according to the present invention scheme, is entered by the attribute according to commodity sales promotion to commodity sales promotion
Row Semantic Clustering such that it is able to ensure to recommend efficiency;It is similar by the way that different equivalence classes are divided into
Equivalence class and related equivalence class such that it is able to targetedly select to recommend in some equivalence classes
Commodity, it is ensured that the reasonability of commercial product recommending;By the demand recommendation grade for calculating related equivalence class,
So as to when selecting commodity to be recommended from related equivalence class, the selection that can try one's best meets
The commodity of consumer demand are recommended;By having selected the two of commodity and unselected commodity in consumer
Under kind of different scenes, recommended by Different Strategies such that it is able to disappearing under different scenes
Expense person recommends to be close to the commodity or grouping of commodities for completely subtracting difference, improves consumer and lives to completely subtracting promotion
Dynamic experience effect.
Specific embodiment
The one exemplary embodiment of the present invention is explained below in conjunction with accompanying drawing, including this
They should be thought only exemplary by the various details of bright embodiment to help understanding.
Therefore, it will be appreciated by those of ordinary skill in the art that, the embodiments described herein can be done
Go out various changes and modifications, without departing from scope and spirit of the present invention.Equally, in order to clear
Chu eliminates the description to known function and structure with concisely in description below.
Fig. 1 be it is according to embodiments of the present invention for the full deactivation of electric business it is dynamic gather single commercial product recommending
The schematic diagram of the key step of method.
As shown in figure 1, the embodiment of the present invention for the full deactivation of electric business it is dynamic gather single commercial product recommending
Method the step of mainly include:
Step S11:Commodity sales promotion is divided into different equivalence classes by way of cluster.
To ensure the reasonability of commercial product recommending, improve and recommend efficiency, first have to enter commodity sales promotion
Row cluster.Carried out by the way of Semantic Clustering in the present invention.Commodity have the attribute of its own,
Such as keyword, brand, price, purposes can be used as the influence factors of Semantic Clustering.Language
Justice cluster adopts K-means algorithms.K-means algorithms receive parameter k;Then will be previously entered
N data object be divided into k cluster so that the cluster that obtained meets:It is same poly-
Object similarity in class is higher;And the object similarity in different clusters is less.Cluster is similar
Degree is to obtain one " center object " (center of attraction) to enter using the average of object in each cluster
What row was calculated.
In one embodiment of this invention, obtain after commodity sales promotion list, calculated using K-means
Method is by commercial articles clustering.Setting similarity threshold α, randomly selects on the basis of commodity G (in cluster
The heart), make D (G, Gn)<α updates cluster centre, when not having as the Aggregation standard of equivalence class
Commodity are chosen next benchmark and are clustered when can enter equivalence class, until not having in promotion list
There are commodity to enter equivalence class, cluster is completed.
Wherein, D is exactly semantic distance, and the semantic distance of only two kinds of commodity is not more than user and sets
During fixed threshold alpha, it was demonstrated that as the two phase, can gather for equivalence class.For example, Yong Huke
, there are two kinds of commodity goods1 and goods2, three attribute in given threshold α=0.8:Price, uses
On the way, keyword (trade name).Goods1 and similarity degrees of the goods2 on keyword are 0.7,
That is d_key (goods1, goods2)=0.7.In the same manner, d_purpose (goods1, goods2)=0.5,
D_price (goods1, goods2)=0.4.User can set weight and be weighted w_price=
0.2, w_purpose=0.3, w_key=0.5.Then:
D (goods1, goods2)=w_price*d_price (goods1, goods2)+w_purpose*d_
Purpose (goods1, goods2)+w_key*d_key (goods1, goods2)=0.58<0.8, therefore this
Two kinds of commodity can not be clustered together.
To meet the characteristic of commodity sales promotion, it is also possible to wherein in a kind of embodiment, choose a certain
Major influence factors of the attribute (such as purposes) as commodity Semantic Clustering.According to commodity sales promotion
Attribute carries out Semantic Clustering, and commodity sales promotion is divided into different equivalence classes.
In concrete operations, commercial articles clustering uses attribute as the influence factor (mark of semantic distance
It is accurate), (SKU is for big chain store dispensing for the commodity sign of commodity in use, such as SKU
One necessary method of center logistics management.Instantly it has been extended to product Unified number
Referred to as, every kind of product is to there is unique No. SKU) deposit in equivalence as the major key for storing
In class.
Step S12:Different equivalence classes are divided into similar equivalence class and related equivalence class, and is counted
Calculate the demand recommendation grade of each related equivalence class.
Different equivalence classes are divided into similar equivalence class and related equivalence class includes:If consumer is
Commodity are selected, then it is similar equivalence class that consumer has selected equivalence class belonging to commodity, remaining etc.
Valency class is related equivalence class;If the non-selected commodity of consumer, the commodity of consumer's purchase in the recent period
Affiliated equivalence class is similar equivalence class, and remaining equivalence class is related equivalence class.
Specifically, under the scene that consumer has selected commodity, system first by consumer
The commodity of selection are clustered with the commodity sales promotion equivalence class after cluster, if it can cluster successfully
This category of equal value is designated as into similar equivalence class, related equivalence class is otherwise labeled as.
Can select the commodity sign of commodity in concrete operations using consumer, such as SKU, and
Whether clustered equivalence class matching, see and select commodity in this equivalence class.
Under the scene of the non-selected commodity of consumer, the SKU of the commodity that consumer is bought in the recent period
Match with clustered equivalence class, see and select commodity whether in this equivalence class.
By said method, just different equivalence classes is divided into similar equivalence class and correlation is of equal value
Class.Similar equivalence class and related equivalence class are distinguished, is to carry out commercial product recommending to consumer
When, targetedly select commodity to be selected the replacement of commodity from similar equivalence class,
Or the high commodity of selection demand recommendation grade are supplemented from related equivalence class.
Under real-world situation, for consumers, it is needed the commodity in different related equivalence classes
Degree is asked to be different, therefore never with selecting commodity to be recommended for disappearing in correlation equivalence class
The purchase experiences of expense person are different.For example, in the case of consumer selects commodity, disappear
Expense person have selected certain commodity, if gathering during list, to its recommended requirements not high yield
Product, then the wish of its purchase is relatively small, and if recommending other high products of consumer demand,
Then its purchase intention is relatively large.Also it is thus, such as in the case of consumer's non-selected commodity
In transaction record before, certain commodity had been bought before consumer, then in the mistake recommended
Still recommend the commodity similar with purchased item in journey, the wish of its purchase is relatively small, and such as
Fruit recommends other products, then purchase intention is relatively large.Therefore, different equivalence classes are divided into
After similar equivalence class and related equivalence class, need the demand for calculating each related equivalence class to recommend etc.
Level.
When consumer selects commodity, the demand recommendation grade of related equivalence class can be according to basis
Consumer has selected the relevance grades of commodity and each related equivalence class and consumer to each correlation etc.
Two factors of inclined happiness degree of valency class, using each related equivalence class of commercial product recommending technology weighted calculation
Demand recommendation grade.
Wherein, consumer has selected commodity to be calculated as follows to the relevance grades of each related equivalence class:
Consumer is selected commodity entered according to item property to the commodity of the cluster centre of related equivalence class
Row Semantic Clustering.The for example aforementioned semantic distance computational methods of method, here is omitted.Draw and disappear
Expense person has selected the semantic distance of commodity and each related equivalence class, and semantic distance is bigger, illustrates consumption
Person has selected commodity equivalence related to this more uncorrelated, then more worth recommendation.Consumer is to each correlation
The inclined happiness degree of equivalence class can be calculated according to consumer's degree of happiness historical data partially, specifically can root
According to big data parser, with the history consumption of consumer, data are browsed as training sample, count
Calculate inclined happiness degree of the consumer to the related equivalence class.For example, in abovementioned steps, to promote
The hierarchical cluster attribute of commodity, then according to consumer degree of happiness historical data analysis consumer partially to such business
The inclined happiness degree that the integrated conditions such as brand, purposes, the price of product consider.The inclined happiness degree is made
For a coefficient, the aforementioned relevance grades of commodity and each related equivalence class that select for calculating are made
For another coefficient, it is weighted using commercial product recommending technology, draws each related equivalence class
Demand recommendation grade.
If the non-selected commodity of consumer, the commodity bought in the recent period according to consumer are related etc. to each
Two factors of inclined happiness degree of the relevance grades and consumer of valency class to each related equivalence class, profit
With the demand recommendation grade of each related equivalence class of commercial product recommending technology weighted calculation.Herein demand is pushed away
Recommend grade sequence method has selected the logic under commodity situation identical with aforementioned consumer, herein no longer
Repeat.
Step S13:According to different Generalization bounds, from similar equivalence class and/or related equivalence class
It is middle to select to recommend completely to subtract the commodity sales promotion or grouping of commodities of difference to fill to consumer, and export
Recommendation results collection.
After making a distinction and sorting to different equivalence classes, according to different Generalization bounds, from
Select to recommend to fill the rush for completely subtracting difference to consumer in similar equivalence class and/or related equivalence class
Pin commodity or grouping of commodities.In the specific embodiment of the invention, in following strategy can be used
It is individual or multiple:
Strategy one:It is close to from recommended price in the high related equivalence class of demand recommendation grade and completely subtracts difference
The commodity or grouping of commodities of volume, completely subtract difference to fill;
Strategy two:Phase is found in the similar equivalence class belonging to a certain commodity that consumer has selected
The a certain commodity selected like described in commodity are substituted, completely subtract difference to fill;
Strategy three:Utilization strategies two select similar commodity to be substituted, when the similar commodity for selecting
But the inclined happiness degree of close consumer price fail to fill when completely subtracting difference, and Utilization strategies one are from needing
Recommended price in the high related equivalence class of recommendation grade is asked to be close to the commodity or commodity group for completely subtracting difference
Close, completely subtract difference to fill.
For example, if consumer has selected commodity, Utilization strategies one in strategy three or
Multiple policy recommendation commodity;If the non-selected commodity of consumer, the Recommendations of Utilization strategies one.
In the selection of specific strategy, if consumer has selected commodity, using strategy one to plan
One or more policy recommendation commodity in slightly three;If the non-selected commodity of consumer, using plan
A slightly Recommendations.
For example, in strategy one, completely subtract that to gather the primary reference point of single activity recommendation be exactly price
Wish that having bought close prices completely subtracts difference and useful commodity with demand, i.e. consumer.Wherein
Demand is i.e. from Recommendations in the high related equivalence class of demand recommendation grade.Therefore, specifically pushing away
During recommending, a set M can be built, as recommendation results collection.Wherein, G is set
In element, each element can be made up of the combination of a commodity or various commodity, then
Recommended using greedy algorithm algorithm:
A commodity in selection demand recommendation grade highest correlation equivalence class add of M
In element, referred to as " currentElement ", and judge the commodity price in the currentElement it
Whether less than difference is completely subtracted, if less than difference is completely subtracted, continuing selection demand recommendation grade most
A commodity add the currentElement in high related equivalence class, repeat current step;
If more than difference is completely subtracted, judging whether the commodity price sum in the currentElement is more than and pushing away
Threshold value is recommended (if as it was previously stated, completely subtract the amount of money for 200, can recommend according to actual conditions setting
The upper limit of commodity price summation, for example, can be 300, so on the one hand avoid recommendation results from exceeding
Consumer is expected to the psychology of pricing for participating in advertising campaign, on the other hand avoids the waste for calculating),
If more than threshold value is recommended, a commodity being eventually adding in the currentElement are deleted, and
Repeat current step;If adding knot less than threshold value, the currentElement commodity is recommended
Beam;
Repeat the above steps, until obtaining containing many that (i.e. n in Fig. 2, herein n can be with root
Define according to actual conditions, represent recommendation results and be concentrated with what n was made up of commodity or grouping of commodities
Element) individual element recommendation results collection.Idiographic flow see shown in Fig. 2.
From the above, it can be seen that being carried out to commodity sales promotion by the attribute according to commodity sales promotion
Semantic Clustering such that it is able to ensure to recommend efficiency;It is similar etc. by the way that different equivalence classes are divided into
Valency class and related equivalence class such that it is able to targetedly select to recommend business in some equivalence classes
Product, it is ensured that the reasonability of commercial product recommending;By the demand recommendation grade for calculating related equivalence class,
So as to when selecting commodity to be recommended from related equivalence class, the selection that can try one's best meets
The commodity of consumer demand are recommended;By having selected the two of commodity and unselected commodity in consumer
Under kind of different scenes, recommended by Different Strategies such that it is able to disappearing under different scenes
Expense person recommends to be close to the commodity or grouping of commodities for completely subtracting difference, improves consumer and lives to completely subtracting promotion
Dynamic experience effect.
Above-mentioned specific embodiment, does not constitute limiting the scope of the invention.This area
Technical staff is it is to be understood that depending on design requirement and other factors, can occur various
The modification of various kinds, combination, sub-portfolio and replacement.It is any within the spirit and principles in the present invention
Modification, equivalent and improvement for being made etc., should be included within the scope of the present invention.