CN106294489A - Content recommendation method, Apparatus and system - Google Patents

Abstract

This application discloses a kind of content recommendation method, device and system.One detailed description of the invention of the method includes: obtain customer attribute information and/or environment attribute information；Content recommendation is synthesized based on customer attribute information and/or environment attribute information.This enforcement can provide more information in the content recommended, and promotes the utilization rate of content recommendation display location.And it is possible to provide more targeted individualized content, thus promote the conversion ratio of content recommendation.

Description

Content recommendation method, Apparatus and system

Technical field

The application relates to field of computer technology, is specifically related to field of terminal technology, particularly relates to Content recommendation method, Apparatus and system.

Background technology

Existing content recommendation system (such as advertisement commending system, information recommendation system etc.) is main Will based on the user detected being carried out interest, hobby, the analysis of Focus Area are recommended to be correlated with Content.This kind of content recommendation system mainly includes three modules: user detects and feature analysis Module, recommending module and display module.

In user's detection with characteristics analysis module, after camera collection to image, use The method such as pedestrian detection, Face datection extracts the targeted customer in image, then to targeted customer Carry out feature analysis.Feature the most frequently used in prior art includes the sex of user, age, clothing The shallow-layer features such as the expression color and style and user.

The recommendation decision-making of recommending module mainly includes two ways.A kind of mode is according to presetting rule Then recommending, such as, be women when sex user being detected, system can be according to presetting Women pay close attention to content and recommend, specifically, cosmetics, fashionable dress can be recommended to women Deng commodity, it is also possible to recommend health, the information of beauty treatment class to women.Another way is to pass through The mode of study is recommended.Can by the feature of user and the characteristic vector of content to be recommended, By machine learning, two kinds of features are associated under Xian, will be used by the model of training on line The Feature Mapping at family is the feature of content to be recommended, with the feature of content to be recommended at content-data Storehouse is mated, thus using content high for matching degree as recommendation results.Wherein, content data base In save preset content, the form of these preset contents and the quantity of information comprised less, and Form is fixing.

In prior art, display module typically only has one piece of display screen, is designed to only serve one Group user.The most only show the content recommended for one group of user.This group is used Family may comprise multiple individuality, by extracting the common characteristic between multiple individualities, such as, extracts Common age level, social relations etc., based on common characteristic content recommendation.

In above-mentioned prior art, there are the following problems:

User's detection with in characteristics analysis module, only the shallow-layer feature of user is analyzed and Extract.In actual applications, it is difficult to judge whether commodity or information fit by these shallow-layer features Conjunction is recommended to user, such as, when detecting that user is young woman, system judges that user can Can be interested in cosmetics, but cannot judge recommend what brand or which kind of class to user The cosmetics of type.

In recommending module, content data base saves preset content, it is recommended that preset content is deposited In following defect: first, it is recommended that module decision-making may go out multiple preset content to be recommended, many Between individual preset content, relatedness may be poor, by showing multiple preset content meetings to be recommended User is made to have fracture sense when the information of acquisition；Second, if user characteristics and content to be recommended Characteristic matching degree is the lowest, then recommendation effect is poor；3rd, that is recommended in prior art is interior Appearance form and the information comprised are all fixing, it is impossible to provide personalized content recommendation, full The demand of foot different user, such as, when recommending wrist-watch, be equipped with different personages and music, user Perception to wrist-watch differs, it is recommended that effect entirely different.

In display module, prior art only serves one group of user, and this mode is difficult to meet The individual demand of multi-user, it is recommended that the quantity of information of content is less.When number of users is too much or special Going on a punitive expedition when complexity, system may completely cannot decision-making content recommendation.

Summary of the invention

For above-mentioned defect of the prior art, it is desired to be able to provide the content of a kind of personalization to push away Recommend method.Further, also it is desirable to the content recommended can have different characteristic for many groups User, comprise more rich information.In view of this, this application provides content recommendation method, Apparatus and system.

First aspect, this application provides a kind of content recommendation method.The method includes: obtain Customer attribute information and/or environment attribute information；Believe based on customer attribute information and/or environment attribute Breath synthesis content recommendation.

In certain embodiments, synthesis content recommendation includes: based on customer attribute information and/or ring Border attribute information determines alternating content element, and wherein alternating content element includes candidate target element With alternate scenes element；And synthesize in recommendation according to candidate target element and alternate scenes element Hold.

Second aspect, this application provides a kind of content recommendation device.This device includes: obtain Unit, is configured to obtain customer attribute information and/or environment attribute information；And synthesis unit, It is configured to synthesize content recommendation based on customer attribute information and/or environment attribute information.

In certain embodiments, synthesis unit includes determining subelement, is configured to based on user Attribute information and/or environment attribute information determine alternating content element, wherein alternating content element bag Include candidate target element and alternate scenes element；And synthon unit, it is configured to according to time Select object elements and alternate scenes element synthesis content recommendation.

The third aspect, this application provides a kind of content recommendation system.This system includes processor And display device；Wherein display device is configured to show content recommendation；Processor includes such as this The content recommendation device of application second aspect.

The content recommendation method of the application offer, Apparatus and system, based on user property and environment Attribute synthesis or generation content recommendation.Can automatically recommend the content of personalization, that is recommended is interior Hold and comprise more information, improve the specific aim of content recommendation system.Meanwhile, can recommend The content meeting user's request and interest that can not directly judge from the shallow-layer feature of user, promotes The utilization rate of content recommendation system.

Accompanying drawing explanation

Non-limiting example is described in detail with reference to what the following drawings was made by reading, this Shen Other features, objects and advantages please will become more apparent upon:

Fig. 1 shows the exemplary flow of the content recommendation method according to one embodiment of the application Figure；

Fig. 2 shows the effect schematic diagram of the ontoanalysis according to the embodiment of the present application；

Fig. 3 shows the exemplary of the determination alternating content element according to one embodiment of the application Flow chart；

Fig. 4 shows the example of the method for the synthesis content recommendation according to one embodiment of the application Property flow chart；

Fig. 5 a shows that the mode using space division to present shows a kind of effect signal of content recommendation Figure；

Fig. 5 b shows that the mode using space division to present shows that the another kind of effect of content recommendation is shown It is intended to；

Fig. 6 shows that the mode using the time-division to present shows the principle schematic of content recommendation；

Fig. 7 shows the structural representation of the content recommendation device according to one embodiment of the application Figure；And

Fig. 8 shows the structural representation of the content recommendation system according to one embodiment of the application Figure.

Detailed description of the invention

With embodiment, the application is described in further detail below in conjunction with the accompanying drawings.It is appreciated that , specific embodiment described herein is used only for explaining related invention, rather than to this Bright restriction.It also should be noted that, for the ease of describe, accompanying drawing illustrate only with About the part that invention is relevant.

It should be noted that in the case of not conflicting, the embodiment in the application and embodiment In feature can be mutually combined.Describe this below with reference to the accompanying drawings and in conjunction with the embodiments in detail Application.

In the following description, a large amount of details are set forth to provide embodiments of the invention Complete description.But, it should be appreciated by those skilled in the art that embodiments herein is not having In the case of having these details, it is also possible to be carried out.

Refer to Fig. 1, it illustrates content recommendation method according to one embodiment of the application Exemplary process diagram.In order to make it easy to understand, in the present embodiment, in conjunction with being used for showing content recommendation Equipment illustrate.It will be understood by those skilled in the art that this is used for showing content recommendation Equipment can be electronic large screen, be such as arranged at underpass or the electrical screen in hall, market Curtain, it is also possible to be such as mobile phone, panel computer etc. the mobile electronic device with display function.

As it is shown in figure 1, in a step 101, customer attribute information and/or environment attribute letter are obtained Breath.

In the present embodiment, user property can be extracted based on the image that photographic head is captured Information.This photographic head can be installed in the equipment for showing content recommendation, it is also possible to quilt It is installed on for showing the one or more positions near the equipment of content recommendation.Realize at some In, the image of multiple camera collection can be obtained, to the user of all users in the range of showing Attribute information is analyzed.

Customer attribute information can include user's individual attribute information and group property information.Wherein, Individual attribute information can be by analyzing the information obtained by the personal feature of each user, Group property information can be based on multiple users between relation obtained by information.

In some implementations, individual attribute information can include the personal feature information of user, all Such as the external appearance characteristic information of user, the information that i.e. can immediately arrive at from the surface character of user, The sex of such as user, age, race, clothing style, cosmetic style, health status, appearance The information such as state.Wherein sex, age, race, cosmetic style can be by the feature of face area Being obtained by grader classification, health status can be by the feature of face area and attitude by dividing Class device or retrieval obtain.Personal feature information can also include the emotional state information of user, example As glad, sad, angry etc., these information can also analyze user's by using grader Facial expression and limb action obtain.

Further, individual attribute information can also include personality and the purchasing power of user.Wherein Personality can include following at least one: sense of responsibility, degree of being emotionally stable, export-oriented degree, to newly The degree of opening of things, affinity, pouplarity, Confidence and lonely degree.These Personality index can be quantified as multiple grade, the corresponding different matching degree of each grade.Example Such as 7 grades from-3 to+3 ,-3 can represent and least meet ,+3 can represent and best suit. As a example by Recommendations, when the degree of opening quantification of targets to new things of user is+3, can To recommend the commodity of some novelties to user, such as, worn the different style of width or a face from user The clothing of color, or recommend travel information；On the contrary, when the degree of opening to new things of user When quantification of targets is-3, the commodity consistent with the holding of user's current state can be recommended to user, Such as with the clothes of style, accessories etc..

User feature analysis can be obtained by the personality of user by returning device.Realize at some In, personality can be obtained by the method for machine learning.Such as can build based on data with existing Vertical training set, uses the method training personality model of machine learning, obtains user characteristics and personality Mapping relations.It is alternatively possible to using the personality index that manually quantifies as training data to property Lattice model is trained.For example, it is possible to obtain the image of multiple user, and extract user characteristics, Be then based on psychological analysis the export-oriented degree of each user is quantified as-3 ,-2 ,-1,0,1, 2,3 totally seven grades ,-3 represent that export-oriented degree are minimum, and+3 represent that export-oriented degree are the highest.Example As user wears bright clothes, the export-oriented degree of this user can be quantified as+2 or +3.Using the export-oriented degree after the feature of user and quantization as training data, use and such as support The export-oriented Degree Models of machine learning method training such as vector machine (SVM), random forest, obtain User characteristics and the mapping relations of export-oriented degree.When judging the export-oriented degree of user, can make It is analyzed with this extroversion Degree Model.In further realizing, can be based on user Multiple features and multiple personality index are trained, then the model obtained is comprehensive lattice model, The judged result of many characters index of user directly can be obtained by this comprehensive lattice model.

Purchasing power information can be from the clothes of user's dress, footwear, the pricing information of accessories worn Middle acquisition.It is possible, firstly, to extract the clothes of user, footwear, the feature of accessories, and data base In make a look up coupling, to obtain clothes, footwear, the brand message of accessories and/or pricing information. It is alternatively possible to calculate pricing information pricing tier in the price of all similar commodity, from And determine the purchasing power of user.For example, it is possible to extract the feature that user is worn a wrist watch, at commodity Data base searches brand message and the pricing information of this wrist-watch.Further, it is also possible to inquiry With the price range information of brand wrist-watch, by pricing information or price range information at all wrist-watches Sequence (such as sequence percentage ratio etc.) in commodity.Alternatively or additionally, purchasing power can be entered Row quantifies, as being quantified as multiple grade.Specifically, if the price of this wrist-watch is at all types Wrist-watch in price be ordered as front 10%, then the purchasing power of user can be defined as highest ranking.

The group property information of user can include the social relations information of user, closes including family System, lovers' relation, friends etc..Can be from the clothes that multiple users are worn to determine use The social relations information at family, such as, can be determined between multiple user by lovers' clothes or parent-offspring's dress be Lovers' relation or family relation.

In some optional implementations, customer attribute information can be obtained in the following way: Gathering the image of the display location region of content recommendation, it is right to determine from image as service The user of elephant, carries out ontoanalysis and group analytic to the user as service object.

In above-mentioned implementation, image acquisition device (such as video camera, movement can be passed through Shooting part in terminal) gather the image of content recommendation display location region.Optional Ground, it is also possible to before gathering image, by sensor detection display location region being The no position that there is user and user, and rotated by computer system control video camera, gather Jiao gathers the image of detected user.

Alternatively, from described image, determine the user as service object, may include that inspection Pedestrian in altimetric image and the sight line focal position of described pedestrian；Judge the sight line focus of pedestrian Whether position is positioned at the display location of content recommendation；If it is, determine that pedestrian is as service The user of object.After collecting image, can come based on features of skin colors or body shape feature Carry out pedestrian detection, it would however also be possible to employ the machine learning such as random forest, HMM Method carries out pedestrian detection, extracts the human body in image.Afterwards can color based on pupil spy Levy (such as black) and shape facility (sub-circular), use such as edge extracting, Hough to become The pupil position of method detection human body such as change, so that it is determined that the location parameter of the sight line focus of pedestrian, And judge whether the sight line focal position of pedestrian in image is positioned at content recommendation based on location parameter Display location.If it is, the pedestrian detected can be defined as the use as service object Family.It should be noted that the display location of content recommendation potentially includes a region, work as pedestrian Sight line focus when being positioned at this region, i.e. it is believed that pedestrian is the user as service object.

In some optional implementations, user carries out ontoanalysis can be by such as lower section Formula is carried out: according to human body, each user in image is divided into multiple subimage, adopts It is analyzed obtaining customer attribute information to subimage with grader and/or recurrence device.Specifically Ground, can carry out image segmentation by the user detected in image according to the different parts of human body. Human body image such as can be divided into face-image, extremity image and body image.Then Can use grader and/or recurrence device that each subimage is analyzed.For example, it is possible to adopt With cosmetic genre classification device, face-image is classified, use clothing genre classification device to extremity Image and body image are classified, thus obtain the multiple attribute information of user.

With further reference to Fig. 2, it illustrates the effect of the ontoanalysis according to the embodiment of the present application Schematic diagram.As in figure 2 it is shown, the user images extracted can be divided into hair image, face Portion's image, left arm image, bag image, left lower limb image, skirt image, left footwear image, right footwear Multiple subimages such as image, right lower limb image, jacket image, right arm image and glasses picture picture. Use grader or recurrence device that each subimage is analyzed, different users can be obtained Attribute information.Such as, hair image carries out classification to obtain the hair style style of user and send out Matter, be analyzed face-image drawing the sex of user, the age, race, expression, The attribute information such as skin, facial characteristics, to the left arm image of user, right arm image, a left side Lower limb image, the analysis of right lower limb image can obtain user strength and health degree etc. other Feature, to bag image, glasses picture picture, left footwear image, right footwear image, skirt image and jacket Image be analyzed obtaining the clothing of user preference and accessories type, the brand of user preference, The attribute informations such as price and collocation commodity.The analysis of glasses picture picture can also be obtained user inclined Good glasses function information.These customer attribute informations all such as can be able to be adopted with quantization means Represent by graduate mode as the aforementioned.

Return Fig. 1, further, user is carried out group analytic and includes described user grouping. A kind of optional implementation is clothing based on users multiple in image, the correlation degree of attitude And relative position information, use grader according to social relations, the multiple users in image to be entered Row classification.Such as can the styles of clothing based on users multiple in image, pattern the most identical Judge whether multiple user is lovers' relation or kinsfolk's relation, it is also possible to according to user's Intimate degree analyzes whether user is friends or lovers' relation.Such as when detecting two When user has extremity, can primarily determine that two users are friends or family relation, Clothing based on two users are the most identical again determines whether lovers' relation.Another kind of optional Implementation is based on ontoanalysis result, utilizes customer attribute information to users multiple in image Cluster.The method of cluster can be by after the attribute information quantization of each user, calculates and uses Distance between the attribute information of family, divides distance less than the customer attribute information of a predetermined threshold value It is one group, and then by corresponding user grouping.Alternatively, when cluster, can preferentially use The group property information (such as social relations) of user clusters, and uses the use of individuality afterwards Family attribute information clusters.

By the acquisition mode of customer attribute information described above, it is possible not only to get richer Rich shallow-layer feature, such as sex, age, race, health status, cosmetic style, accessories wind Lattice etc., it is also possible to get the further feature of user, such as personality, purchasing power etc., thus can To recommend more to meet user's request or user's content interested to user based on these features, And then the conversion ratio of content recommendation can be improved.

In certain embodiments, the mode obtaining environment attribute information can include but not limited to lead to Cross network receive current temporal information and/or receive the displaying position with content recommendation by network Put corresponding spatial information.Wherein temporal information can at least include with the next item down: current day Time phase, weather condition, festival information, current hot ticket.Spatial information can include exhibition Show in the geographic orientation of position and/or the terrestrial reference of adjacent domain, such as airport, waiting room, business The heart etc..

In the embodiment of the present application in addition to obtaining user property, it is also possible to environment attribute is carried out Analyze and obtain, utilize environment attribute that the content recommended is analyzed decision-making, it is provided that More meet the content recommendation of ambient condition, improve the ageing of commending contents.

In a step 102, synthesize in recommendation based on customer attribute information and/or environment attribute information Hold.

In the many embodiments of the application, content is divided into multiple element, between these elements Various combination can be carried out, thus produce different contents.In this way, commending contents System, without prestoring substantial amounts of immobilized substance, only need to store various content element, it is possible to Generate abundant content, thus the content of this change is referred to as dynamic content.

In certain embodiments, step 102 can include step 1021: believes based on user property Breath and/or environment attribute information determine alternating content element.

In the present embodiment, alternating content element can include candidate target element and alternate scenes Element.In some optional implementations, object elements can be commodity, and situation elements can Think ad elements.Correspondingly, candidate target can be candidate quotient product, and alternate scenes element can Think candidate locations element.Object elements and situation elements can have many attribute.Object meta The attribute of element can include but not limited in the classification of commodity, price, color and brand at least One, the attribute of situation elements can include but not limited to visual style, plot, be suitable for Commodity, personage, time, place and at least one in dubbing in background music.

In some implementations, can be based on the customer attribute information acquired in step 101 and/or ring Border attribute information, utilizes that recommended models collection is incompatible determines candidate's element content.Recommended models set The first recommendation according to customer attribute information recommended element and/or situation elements can be included Model set, object elements and/or situation elements by the second recommended models set of combine recommendation, According to environment attribute information recommendation object elements and/or the 3rd recommended models set of situation elements In at least one.

In further realizing, the first recommended models set can be that customer attribute information is with right As the set of the submodel of interest relation between the attribute of attribute of an element and/or situation elements, wherein The attribute of at least one customer attribute information and object elements and/or situation elements can be included The submodel of interest relation between attribute.Second recommended models set can be the attribute of object elements And/or the set of the submodel of interest relation between the attribute of situation elements, wherein can include at least The submodel of interest relation between the attribute of one object elements and/or the attribute of situation elements.3rd Recommended models set can be environment attribute information and the attribute of object elements and/or situation elements Attribute between the set of submodel of interest relation, wherein can include at least one environment attribute The submodel of interest relation between the attribute of information and the attribute of object elements and/or situation elements.

With object elements as commodity, as a example by situation elements is ad elements, the first recommended models collection Each submodel in conjunction can represent according to user property each commodity or each advertisement Element carries out the mapping relations recommended；Each submodel in second recommended models set can be with table Show different commodity/ad elements by the mapping relations of combine recommendation, in the 3rd recommended models set Each submodel can represent to enter each commodity or each ad elements according to environment attribute The mapping relations that row is recommended.

In some optional implementations of the present embodiment, after determining recommended models set, permissible Use at least one submodel pair in recommended models set relevant to user property and environment attribute Content recommend.

Refer to Fig. 3, it illustrates the determination alternating content unit according to one embodiment of the application The exemplary process diagram of element.In the embodiment that Fig. 3 is corresponding, recommended models set is utilized to determine The method of alternating content element may include that

Step 301, is carried out the submodel in recommended models set based on interest-degree statistical data Training, to determine the parameter of submodel.

As it has been described above, each recommended models set can include one group of submodel, submodule Type can represent user property and object elements, user property and situation elements, different object meta Between element, between different situation elements, environment attribute and object elements, environment attribute and scene The mapping relations of element.In the present embodiment, this mapping relations can be by interest-degree statistical number According to drawing.Specifically, can draw based on interest-degree each submodel of statistical data training The parameter of submodel.Wherein interest-degree statistical data may include that customer attribute information is to object The interest-degree statistical data of the attribute of attribute of an element and/or situation elements, different object elements Interest-degree statistical data between attribute, the interest-degree statistics between the attribute of different situation elements Data, the interest-degree statistical data between attribute and the attribute of situation elements of object elements, with And the interest-degree system that environment attribute information is to the attribute of object elements and/or the attribute of situation elements Count.

Interest-degree statistical data can be quantified as multiple grade, it is also possible to is quantified as normalized Numerical value.The mode obtained can be that the data statistics by online shopping site obtains, such as Can add up and purchase the pageview of a certain commodity, purchase volume and browse or buy this on net website online The corresponding relation of age bracket belonging to the user of commodity, thus add up different age group user to this business The interest-degree statistical data of product.The most such as can be by buying the number of users of multiple commodity simultaneously (the most simultaneously buying the number of users of the refrigerator of certain brand and the washing machine of another brand) statistics is come Calculate refrigerator and the interest-degree statistical data of washing machine, refrigerator brand and the interest of washing machine brand Degree statistical data.The another kind of acquisition mode of interest-degree statistical data is carried out for questionnaire by inquiry Statistics.Such as can design questionnaire targetedly, add up all ages and classes, personality, purchase The user of the power interest-degree to different Brand.Can also be set by experience, the most permissible Empirically set the women interest-degree statistical data to cosmetics, and be normalized to 0.8, and male Property can be set as 0.2 to the interest-degree statistical data of cosmetics.

Table 1 shows and provides the age genus to object elements in customer attribute information with tabular form One example of the interest-degree statistical data of the brand in property.Wherein interest-degree is normalized to 0 To 1.

Table one age interest-degree statistics table to brand

	Disney	Gap	Eland	……
					0-3	0.8	0.6	0.0	……
3-5	0.6	0.5	0.0	……
					5-10	0.8	0.6	0.3	……
10-20	0.3	0.8	0.8	……
					……	……	……	……	……

As can be seen from Table 1, the interest-degree statistics table of brand has been added up each age by the age User's interest-degree to Brand, it is likewise possible to add up other customer attribute informations with Interest-degree between interest-degree between different item property or ad elements, different commodity, no With the interest-degree of ad elements and environment attribute information to different commodity or different ad elements Interest-degree.

Alternatively, in order to the submodel after making training is adapted to environmental change, can be based on newly Object elements and situation elements corresponding submodel is updated.Such as can be according to new product The commodity of board update the interest-degree statistical data relevant to brand and submodel.It addition, also may be used Interest-degree statistical data to be updated with cycle regular hour, use the interest-degree updated The submodel that statistical data training is corresponding, obtains the submodel updated.Such as can quarterly root Submodel is updated according to the feedback information of businessman.

Step 302, sets up global energy function based on recommended models set.

After training obtains the submodel in recommended models set, can be from object elements data Storehouse and situation elements data base are searched according to certain rule meet demand object elements and Situation elements.Can push away based on the first recommended models set, the second recommended models set and the 3rd Recommend model set and set up following function, and determine object elements and scene unit based on formula (1) Element.

productSet^*=argmin_productSetE(productSet|models,userSet,context) (1)

Wherein, productSet represents the set of object elements or situation elements, productSet* Representing the candidate target element or the set of alternate scenes element determined, models represents recommendation mould Type set, models={model₁,model₂,model₃, wherein model₁Represent the first recommendation mould Type set, model₂Represent the second recommended models set, model₃Represent the 3rd recommended models collection Close.UserSet represents the set of customer attribute information, and context represents environment attribute information, E () Represent global energy function.

Determine content recommendation i.e. choose from data base have least energy function object elements and / or situation elements.Global energy function can define such as formula (2):

$\begin{array}{c}E\left(\mathrm{productSet}\right|\mathrm{mode}\mathrm{ls},\mathrm{userSet},\mathrm{context})\\ ={\mathrm{\α}}_1{E}_1+{\mathrm{\α}}_2{E}_2+{\mathrm{\α}}_3{E}_3\\ ={\mathrm{\α}}_1\underset{i,j}{\mathrm{\Σ}}{E}_1\left({\mathrm{product}}_j\right|{\mathrm{mode}l}_1,{\mathrm{user}}_i)\\ +{\mathrm{\α}}_2\underset{j_1,j_2}{\mathrm{\Σ}}{E}_2({\mathrm{product}}_{j_1},{\mathrm{product}}_{j_2}|{\mathrm{mode}l}_2)\\ ={\mathrm{\α}}_3\underset{j}{\mathrm{\Σ}}{E}_3\left({\mathrm{product}}_j\right|{\mathrm{mode}l}_3,\mathrm{context})\end{array}---\left(2\right)$

In formula (2), productSet={product_j, userSet={user_i, wherein i, j, j₁, j₂For positive integer, product_j, product_j1, product_j2Represent object elements or situation elements, user_iRepresent customer attribute information.α₁, α₂, α₃Represent weight coefficient, can rule of thumb set Fixed or training draws.

As shown in formula (2), global energy function can include the first energy function E₁(·)、 Second energy function E₂() and the 3rd energy function E₃(·).First energy function can be basis The energy function that object elements or situation elements are recommended by customer attribute information, specifically, First energy function can calculate according to formula (3):

$E_1\left({\mathrm{product}}_j\right|{\mathrm{mode}l}_1,{\mathrm{user}}_i)=\underset{(p,q)}{\mathrm{\Σ}}{\mathrm{\β}}_{(p,q)}E\left({\mathrm{product}}_j{\mathrm{profile}}_p\right|{\mathrm{mode}l}_1,{\mathrm{user}}_i{\mathrm{profile}}_q)---\left(3\right)$

Wherein, i, j, p, q are positive integer, product_jprofile_pRepresent jth object elements Pth the attribute of/situation elements, user_iprofile_qRepresent the q-th attribute of i-th user, β_{(p, q)}Represent weight coefficient.First energy function may include that based on the first recommended models collection Close, use grader and/or return the object elements corresponding with customer attribute information that device calculates Attribute and/or the recommendation probability of attribute of situation elements.

Second energy function can different object elements or different situation elements be recommended jointly Energy function, specifically, the second energy function can calculate according to formula (4):

$E_2({\mathrm{product}}_{j_1},{\mathrm{product}}_{j_2}|{\mathrm{mode}l}_2)=\underset{(p,q)}{\mathrm{\Σ}}{\mathrm{\λ}}_{(p,q)}E({\mathrm{product}}_{j_1}{\mathrm{profile}}_p,{\mathrm{product}}_{j_2}{\mathrm{profile}}_q|{\mathrm{mode}l}_2)---\left(4\right)$

Wherein, j₁, j₂, p, q are positive integer, product_j1profile_pRepresent jth₁Individual object Pth the attribute of element/situation elements, product_j2profile_qRepresent jth₂Individual object elements/ The q-th attribute of situation elements, β_{(p, q)}Represent weight coefficient.Second energy function can wrap Include: based on the second recommended models set, use grader and/or return the object meta that device calculates The attribute of element and/or the attribute of situation elements are by the probability of combine recommendation.

3rd energy function can be to enter object elements or situation elements according to environment attribute information The energy function that row is recommended, specifically, the 3rd energy function can calculate according to formula (5):

$E_3\left({\mathrm{product}}_j\right|{\mathrm{mode}l}_1,{\mathrm{context}}_i)=\underset{(p,q)}{\mathrm{\Σ}}{\mathrm{\γ}}_{(p,q)}E\left({\mathrm{product}}_j{\mathrm{profile}}_p\right|{\mathrm{mode}l}_3,\mathrm{contextpro}{\mathrm{file}}_q)---\left(5\right)$

Wherein, i, j, p, q are positive integer, product_jprofile_pRepresent jth object elements Pth the attribute of/situation elements, contextprofile_qRepresent the q in environment attribute information Individual attribute, γ_{(p, q)}Represent weight coefficient.3rd energy function may include that and pushes away based on the 3rd Recommend model set, use grader and/or return that device calculates corresponding with customer attribute information The recommendation probability of the attribute of object elements and/or the attribute of situation elements.

Continue Fig. 3, in step 303, global energy function is carried out global optimization and solves, obtain Make the alternating content element of global energy Function Optimization.

In the present embodiment, content recommendation can be determined based on above-mentioned energy function.Specifically, According to formula (1), global energy function can be carried out global optimization to solve.The method of global optimization Optimized algorithm based on genetic algorithm, linear programming, simulated annealing etc. can be included.Solve formula (1), after the productSet* in, i.e. can get candidate target element and alternate scenes element.

In certain embodiments, can be based on the first energy function, the second energy function and the 3rd One in energy function determines alternating content element, such as can be based on the first energy function Recommend wedding gauze kerchief dress ornament to lovers, recommend colorful gorgeous dress ornament to young export-oriented women；Permissible Based on the second energy function by refrigerator and television set, lipstick and eyebrow pencil, infanette and feeding bottle difference Commodity as combine recommendation；Plumage can also be recommended in snowy winter based on the 3rd energy function Floss takes commodity, recommends travel information on the billboard on airport.In some implementations, Ke Yijie Close two or three energy in the first energy function, the second energy function and the 3rd energy function Function determines content recommendation.Such as can recommend lovers' T-shirt in summer to lovers, winter to Lovers recommend lovers' down jackets, recommend lovers' wrist-watch and lovers' ring etc. to lovers simultaneously.

The embodiment illustrated above in association with Fig. 3 describes based on customer attribute information and environment attribute Information determines a kind of method of alternating content element, in actual applications, when content recommendation is wide During announcement, carry out global energy function after global optimization solves, one group to be obtained and preferentially combining The commodity set recommended and ad elements set.

In the method for the determination alternating content element that this example provides, can be according to the interest of user Degree and tendency and/or environmental information select multiple object elements and the situation elements recommended, it is possible to carry For more rich content recommendation.Such as when recommended advertisements, it can be deduced that multiple meet user's request Ad elements and situation elements with hobby so that ad content is abundant, improve making of billboard By rate and input effect.Further, it is possible to provide more lively ad elements, promote user's body Test.

Returning Fig. 1, step 102 may further include step 1022, according to candidate target unit Element and alternate scenes element synthesize content recommendation.

In the present embodiment, step 1021 determine candidate target element and alternate scenes element it After, candidate target element can be merged, alternate scenes element is combined simultaneously, And combine candidate target element and alternate scenes Element generation content recommendation.Can be first by each Individual candidate target element blends with corresponding alternate scenes element, afterwards by multiple alternate scenes Element is combined.

In some implementations, candidate target element and candidate field can be merged based on default rule Scape element.With further reference to Fig. 4, it illustrates the synthesis according to one embodiment of the application and push away Recommend the exemplary process diagram of the method for content.

As shown in Figure 4, in step 401, the placement index of alternate scenes element, side are obtained To index and movement locus index.

In the present embodiment, situation elements typically has transparent background or specific placement location For placing objects element.Can set up in these specific positions and place index, be used for setting The type of the object elements that ad-hoc location can be placed.Such as can place vehicle, hands on road Wrist-watch can be placed at wrist.Further, ad-hoc location can also be set up direction index, use In denoted object element towards.For example, it is possible to determine the placement of vehicle according to the direction of road Towards.Upper arm attitude according to people determine wrist-watch towards.Further, when alternate scenes unit Element is dynamic element, such as during video, it is also possible to set up movement locus index, right to indicate The direction of motion of picture dot element and route.Road scene such as can including, road direction indexes, So that vehicle travels along road direction.

Before candidate target element and alternate scenes element being merged, can first obtain candidate The above-mentioned index information of situation elements, including placing index, direction index and movement locus rope Draw.The mode obtained can be directly to search related data from data base, it is also possible to for field Scape element carries out graphical analysis, video analysis, extracts wherein for placing candidate target element Feature, is determined that by the model of training the placement of this scene indexes, extracts the position in situation elements Put feature, direction character and movement locus feature, thus obtain direction index and movement locus rope Draw.

In step 402, will time according to placement index, direction index and movement locus index Object elements is selected to merge with alternate scenes element, to generate Candidate Recommendation content.

When synthesizing alternating content element, can first candidate target element be put according to placing index Put the ad-hoc location in alternate scenes element, then index candidate target element according to direction Rotate, afterwards according to the mobile candidate target element of movement locus index, synthesize complete time Select content recommendation.

In step 403, based on the degree of association between alternate scenes attribute of an element, candidate target Degree of association between attribute of an element and between alternate scenes attribute of an element and candidate target element Candidate Recommendation content is screened by degree of association, will melt through the Candidate Recommendation content of screening Close, to generate content recommendation.

In the present embodiment, between multiple Candidate Recommendation contents, it is likely to be of certain relatedness, Such as temporal associativity, spatial correlation, personage's relatedness, event correlation and Attribute Association Property.Can be according to these relatednesss screening stronger multiple Candidate Recommendation contents of relatedness, will be with The unrelated Candidate Recommendation content of other Candidate Recommendation contents filters, in the Candidate Recommendation that will filter out Hold and be fused to smoothness, coherent content recommendation.

Relatedness between Candidate Recommendation content can be to be comprised not based on Candidate Recommendation content With between candidate target attribute of an element, candidate target attribute of an element and alternate scenes element Relatedness between attribute and between different alternate scenes attribute of an element determines.Therefore, In the present embodiment, can be according to the relatedness between different candidate target attributes of an element, time Select between the attribute of object elements and alternate scenes attribute of an element and different alternate scenes unit Relatedness between the attribute of element is screened and is comprised corresponding candidate target element or alternate scenes element Candidate Recommendation content.

Relatedness between above-mentioned different candidate target attribute of an element, the genus of candidate target element Property between alternate scenes attribute of an element and between different alternate scenes attribute of an element Relatedness can be obtained by the method for model training, it is also possible to the most manually sets.Close The expression of connection property can be the numerical value quantified.It is alternatively possible to by candidate target attribute of an element Carry out vectorization with alternate scenes attribute of an element, then calculate the distance between an attribute, away from From the least, relatedness is the strongest.Calculate the genus of the object elements that all Candidate Recommendation contents are comprised Relatedness between the attribute of property and situation elements, can filter according to relatedness and push away with other candidates Recommend relevance the most weak or there is no the Candidate Recommendation content of relatedness.

After screening, multiple Candidate Recommendation contents that relatedness is stronger can be obtained, and according to time Between the Candidate Recommendation content series connection that will filter out of order, spatial relation or state-event, raw The content recommendation become.

It is appreciated that if step 1021 only draws a Candidate Recommendation content, or multiple time Select and between content recommendation, all there is no relatedness, then can be using a Candidate Recommendation content as recommendation Content.

For example, when recommended advertisements, the relevant of commodity element and ad elements can be calculated Property, such as two the ad elements dependencys with similar video style are relatively strong, then can will wrap Advertisement containing the two ad elements is positioned over in advertisement first.The most such as two ad elements Scene be Same Scene, time attribute is respectively the morning and noon, the commodity comprised be respectively Car and wrist-watch, then can be first by vehicle advertisement and the advertisement that the time attribute of ad elements is the morning The advertisement that time attribute is noon of element is together in series, and forms the video ads of Time Continuous first.

It should be noted that the example of the method for the synthesis content recommendation of above-mentioned combination Fig. 4 explanation Property realize, screening step can be first carried out, relevant based between alternate scenes attribute of an element Degree, degree of association between candidate target attribute of an element and alternate scenes attribute of an element and candidate Candidate target element and alternate scenes element are screened, then by the degree of association between object elements Obtain the placement index of alternate scenes element, direction index and movement locus index, afterwards by phase Candidate target element that pass degree is high and alternate scenes element are fused to Candidate Recommendation content, and according to Candidate Recommendation content is together in series by the time attribute of alternate scenes element, forms complete recommendation Content.

The content recommendation method that the above embodiments of the present application are provided, based on customer attribute information and Environment attribute information determines candidate target element and alternate scenes element, then according to candidate target Element and alternate scenes element synthesis content recommendation.Can provide more in the content recommended Information, promotes the utilization rate of content recommendation display location.And it is possible to provide more targeted Individualized content, thus promote the conversion ratio of content recommendation.

The method that above-described embodiment is provided may be used for intelligent advertisement commending system.System is permissible Obtain the image before billboard by the photographic head being arranged on billboard, image is carried out human body Detection, detects targeted customer A, just can determine targeted customer A by focus detection afterwards In the content paid close attention on billboard.System can carry out ontoanalysis to targeted customer A, analyzes Result is male, 40-50 year, the high-grade navy blue western-style clothes of dress and black leather shoes, facial characteristics divides Analyse its personality be responsible, self-confident, degree of being emotionally stable by force, more introversive, purchasing power strong, then Black commercial affairs car, dark high-grade POLO shirt, certain brand can be recommended high to this targeted customer The commodity of shelves wrist-watch, it is recommended that ad elements can include classicism music, high-grade household field Scape, commercial affairs office building office scene, city road conditions etc..Finally according to these commodity and advertisement unit Relatedness between element merges, and the advertisement of generation can be that hero wears blueness early morning High-grade POLO shirt, wear high-grade wrist-watch, drive black commercial affairs car regarding to commercial affairs office building Frequently.Period can also be interted hero and be seen and participate in meeting, the feelings left of driving in the setting sun after table Joint.

This intelligent advertisement commending system can also organize recommended advertisements for the user comprising multiple user. Such as 6 targeted customers A, B, C, D, E, F can be detected by analyzing image, Determine that 6 targeted customers are paying close attention to the content on billboard by focus detection.System is permissible First 6 targeted customers are carried out ontoanalysis, may then based on ontoanalysis result and 6 Attitude, relative position relation between individual targeted customer carry out group analytic.Analysis result be A, B, C are family of three, organize 1 as user, and D, E are that the probability of lovers' relation is higher, Organizing 2 as user, F organizes 3 as user.System can generate three sections of advertisements, the most right User organizes 1,2,3 and carries out advertisement recommendation.

In certain embodiments, foregoing recommends method to include:

Step 103, uses the mode that the time-division presents or space division presents to show content recommendation.

If the user as service object obtained in step 101 or user organize quantity for many Individual, after generating content recommendation, need by the way of appropriate to many as service object Individual user or user organize displaying content recommendation.As a example by showing content recommendation on the electronic display screen, In the present embodiment, content recommendation can be shown in the way of the employing time-division present or space division presents. The mode that wherein time-division presents is applicable to the electronic curtain that screen area is less, the side that space division presents Formula is applicable to the bigger screen of screen area or curve screens.

In some implementations, the mode using space division to present shows that content recommendation can be by as follows Mode is carried out: first the display location of content recommendation is divided into and organizes quantity phase with user or user Deng subregion, then would correspond to each user or each group of user content recommendation show In the subregion at this user/place, this group user's sight line focal position.It should be noted that Facial image can be carried out pupil position detection and the degree of depth inspection when carrying out user's sight line focus detection Survey, thereby determine that the position on screen, region that user pays close attention to.Further, it is also possible to root User's field range is determined according to the pupil position of user, so that it is determined that the content recommendation shown Size.

In further realizing, it is also possible to follow the tracks of the sight line focal position of user, according to user The change of sight line focal position, adjust the displaying of the content recommendation of each or each group of user Position.If user or user's group are kept in motion, then can follow the tracks of it by pedestrian detection Change in location, or when user or user's group remain static, but region-of-interest changes Time, the change determining user's sight line focal position can be detected in real time based on pupil position, from And obtain the variable condition of the position that user is paid close attention to.At this moment, the exhibition of content recommendation can be adjusted Show position, make the content organizing recommendation for this user or user can project user or user all the time Group is within sweep of the eye.

With further reference to Fig. 5 a, it illustrates the mode using space division to present and show content recommendation A kind of effect schematic diagram.The scene of Fig. 5 a can utilize hall, market or the billboard in hall, hotel To customer recommendation advertisement.In fig 5 a, for showing that the display screen of content recommendation is cylinder screen Curtain 510.System detects destination service object user 501 and 502, wherein the regarding of user 501 Line focus is positioned in region 511, and the sight line focus of user 502 is positioned in region 512.Pass through User 501 and user 502 are carried out ontoanalysis, draws the user 501 interest-degree to wrist-watch Relatively big with demand degree, user 502 is relatively big to the demand degree of car, then can be respectively in region 511 Watch advertisement and car ads is shown, it is achieved that at the diverse location of cylindrical screen in 512 Different personalized advertisements is recommended for different user.

With further reference to Fig. 5 b, it illustrates the mode using space division to present and show content recommendation Another kind of effect schematic diagram.Display location shown in Fig. 5 b can be that the transfer of ferrum similarly is led to Flat-faced screen in road.These display screens can be laid along wall, and user is in transfer process In, the display screen on wall can present personalized advertisement.As shown in Figure 5 b, screen 520 Many sub regions can be divided into.The sight line of user 503, user 504 and user 505 is burnt Point is in region 521, region 522 and region 523.Can open up on corresponding region Show the content recommended to each user.Such as to user 503 recommend jacket and the advertisement of skirt, Recommend watch advertisement to user 504, recommend car ads to user 505.And it is permissible Real-time tracking user sight line focal position in user's moving process, according to user's sight line focus position The change put adjusts the position of the content recommendation shown.Such as when the sight line focal position of user When moving to region 522, the content that region 522 is shown can be switched to car ads. When user's overlap, can show on screen that sight line is not obstructed the content recommendation corresponding to user.

The screen presented for the time-division can be grating display screen, and quick by grating is moved, Switch different content recommendations.

In some implementations, the mode that the employing time-division presents shows that described content recommendation can pass through The display location of content recommendation switches at a certain time interval corresponding to each The content recommendation of individual or each group of user is carried out.This time interval can be that the vision of human eye is temporary Stay the time, as such, it is possible to utilize the visual persistence phenomenon of human eye to realize the exhibition of multiple content recommendation Show.

In further realizing, it is also possible to follow the tracks of the sight line focal position of described user, according to The change of the sight line focal position of described user, adjusts in each or the recommendation of each group of user The displaying angle held.While showing content recommendation, can be examined in real time by focus detection Survey position and the depth information of user of focus, so that it is determined that the change of the field range of user, The change that may then based on user's field range adjusts the direction of grating, makes content recommendation all the time It is illustrated in user within sweep of the eye.

With further reference to Fig. 6, it illustrates the mode using the time-division to present and show content recommendation Principle schematic.As shown in Figure 6,602 is the video camera for gathering image, light-emitting diodes Image or video are presented to use by pipe (LED) projected array 601 by grating display screen 603 Family, wherein when grating moves to a certain position, left eye and right eye lay respectively at region 611 He The user in region 612 is it can be seen that be included as the First Kind Graph picture of the content that this user recommends；When When raster transform is to another location, left eye and right eye lay respectively at region 613 and region 614 User is it can be seen that be included as the Equations of The Second Kind image of the content that this user recommends.Video camera 602 can With the change of the sight line focal position of detection user in real time, grating is adjusted with the movement of user's sight line Whole angle, it is ensured that the content recommendation of displaying is always positioned at user within sweep of the eye.

The method showing content recommendation above by space division or time-division presentation mode, can be to multiple Or how group user shows personalized multiple content recommendations, improves the displaying of content recommendation simultaneously The utilization rate of position, and can automatically adjust display location by focus detection so that content Recommend more intelligent.

With further reference to Fig. 7, it illustrates the commending contents dress according to one embodiment of the application The structural representation put.

As it is shown in fig. 7, content recommendation device 700 can include acquiring unit 701 and synthesis Unit 702.Acquiring unit 701 may be configured to obtain customer attribute information and/or environment belongs to Property information.Synthesis unit 702 may be configured to based on customer attribute information and/or environment attribute Information synthesis content recommendation.In certain embodiments, synthesis unit 702 can include determining that son Unit 7021 and synthon unit 7022.Determine that subelement 7021 is configured to based on obtaining single Unit's customer attribute information acquired in 701 and/or environment attribute information determine alternating content element, Wherein alternating content element can include candidate target element and alternate scenes element.Synthon list Unit 7022 may be configured to according to determine candidate target element determined by subelement 7021 and Alternate scenes element synthesis content recommendation.

In the present embodiment, acquiring unit 701 can come based on the image that photographic head is captured Extract customer attribute information.Customer attribute information can include user's individual attribute information and colony Attribute information.Wherein, individual attribute information can be by analyzing the individual special of each user Levy obtained information, the sex of user, age, race, clothing style, change can be included The information such as adornment style, health status, attitude.Group property information can be based on multiple users Between relation obtained by information, can be the social relations information of user, including family relation, Lovers' relation, friends etc..In some implementations, customer attribute information can be by multiple Grader classification or recurrence device retrieval obtain.

In some implementations, to be configured to the mode of environment attribute information permissible for acquiring unit 701 Include but not limited to receive current temporal information by network and/or received by network and push away Recommend the corresponding spatial information of display location of content.

In some implementations, determine that subelement 7021 can be based on acquired in acquiring unit 701 Customer attribute information and environment attribute information, utilize recommended models set to build global energy function, Global energy function is carried out optimization to determine candidate's element content.Recommended models set The first recommendation according to customer attribute information recommended element and/or situation elements can be included Model set, object elements and/or situation elements by the second recommended models set of combine recommendation, According to environment attribute information recommendation object elements and/or the 3rd recommended models set of situation elements In at least one.

In further realizing, the first recommended models set can be that customer attribute information is with right As the set of the submodel of interest relation between the attribute of attribute of an element and/or situation elements, wherein The attribute of at least one customer attribute information and object elements and/or situation elements can be included The submodel of interest relation between attribute.Second recommended models set can be the attribute of object elements And/or the set of the submodel of interest relation between the attribute of situation elements, wherein can include at least The submodel of interest relation between the attribute of one object elements and/or the attribute of situation elements.3rd Recommended models set can be environment attribute information and the attribute of object elements and/or situation elements Attribute between the set of submodel of interest relation, wherein can include at least one environment attribute The submodel of interest relation between the attribute of information and the attribute of object elements and/or situation elements.On State interest relation to be represented by interest-degree statistical data.The acquisition of interest-degree statistical data Mode includes but not limited to: experience sets, questionnaire is added up and online shopping site data Statistics.

In some implementations, synthon unit 7022 can obtain the placement rope of alternate scenes element Draw, direction indexes and movement locus index, according to index by candidate target element and candidate field Scape element merges, and generates Candidate Recommendation content, afterwards can be according to candidate target element and candidate Relatedness between situation elements, the relatedness between different alternate scenes element and different candidate couple Relatedness between picture dot element is to Candidate Recommendation Content Selection, in the Candidate Recommendation that finally will filter out Hold and be together in series sequentially in time, form complete, smooth content recommendation.

In certain embodiments, content recommendation device can also include display unit 703.Present Unit 703 may be configured to use the mode that the time-division presents or space division presents to show synthesis unit Content recommendation synthesized by 702.Wherein time-division presentation mode can be provided with removable slit Display screen on present content recommendation, by raster transform angle, utilize the persistence of vision of human eye, The content recommendation organized corresponding to multiple users or user is switched within the retentivity time of eye.Space division in Existing mode can present content recommendation on larger area screen or curve screens, is divided by screen For many sub regions, it is right to show in the subregion that each user or each group of user are paid close attention to The content recommendation answered.In further realizing, it is also possible to follow the tracks of user or the realization of user's group Focal variation, adjusts display location or the angle of content recommendation in real time.

The content recommendation device that the above embodiments of the present application are provided, it is provided that more targeted Individualized content, and provide more information in the content recommended, promote content recommendation The utilization rate of display location and the conversion ratio of content recommendation.

Should be appreciated that what all elements reference Fig. 1-6 described in content recommendation device 700 described Each step in method is corresponding.Thus, the operation described above with respect to method and feature are same Sample is applicable to content recommendation device 700 and the unit wherein comprised, and does not repeats them here.

With further reference to Fig. 8, it illustrates the commending contents system according to one embodiment of the application The structural representation of system.

Content recommendation system 800 at least can include processor 801 and display device 802.Its Middle processor 801 can include the content recommendation device 700 that above-mentioned combination Fig. 7 describes.Display Equipment may be configured to the content recommendation that video-stream processor is generated.It is appreciated that processor Can be independent processing unit, be used for performing content recommendation method.In some implementations, interior Hold commending system and can also include the input equipment of such as keyboard, mouse etc.；Such as hard disk etc. Memorizer, is used for storing candidate target element and alternate scenes element；Such as LAN card, modulation The communication unit of the NIC of demodulator etc., performs communication via the network of such as the Internet Process；And the detachable media of such as disk, CD, magneto-optic disk, semiconductor memory, So that the computer program read from it is mounted into memorizer as required.

As on the other hand, present invention also provides a kind of computer-readable recording medium, this meter Calculation machine readable storage medium storing program for executing can be that computer included in device described in above-described embodiment can Read storage medium；Can also be individualism, be unkitted the computer-readable allocating in terminal unit Storage medium.This computer-readable recording medium storage has one or more than one program, should Program could be included for the program code of the method shown in flow chart that performs.

Flow chart in accompanying drawing and block diagram, it is illustrated that according to various embodiments of the invention system, Device, architectural framework in the cards, function and the operation of method and computer program product. In this, each square frame in flow chart or block diagram can represent a module, program segment, Or a part for code, a part for described module, program segment or code comprises one or many The executable instruction of the individual logic function for realizing regulation.It should also be noted that in some conduct In the realization replaced, the function that marked in square frame can also be marked in accompanying drawing to be different from Order occurs.Such as, two square frames succeedingly represented can essentially perform substantially in parallel, They can also perform sometimes in the opposite order, and this is depending on involved function.Also to note Side in meaning, each square frame in block diagram and/or flow chart and block diagram and/or flow chart The combination of frame, can come by the special hardware based system of the function or operation that perform regulation Realize, or can realize with the combination of specialized hardware with computer instruction.

Above description is only the preferred embodiment of the application and saying institute's application technology principle Bright.It will be appreciated by those skilled in the art that invention scope involved in the application, do not limit In the technical scheme of the particular combination of above-mentioned technical characteristic, also should contain simultaneously without departing from In the case of described inventive concept, above-mentioned technical characteristic or its equivalent feature carry out combination in any And other technical scheme formed.Such as features described above and (but not limited to) disclosed herein The technical characteristic with similar functions is replaced mutually and the technical scheme that formed.

Claims (21)

1. a content recommendation method, it is characterised in that described method includes:

Obtain customer attribute information and/or environment attribute information；

Content recommendation is synthesized based on described customer attribute information and/or described environment attribute information.

Method the most according to claim 1, it is characterised in that described synthesis content recommendation Including:

Alternating content unit is determined based on described customer attribute information and/or described environment attribute information Element, described alternating content element includes candidate target element and alternate scenes element；And

Content recommendation is synthesized according to described candidate target element and alternate scenes element.

Method the most according to claim 2, it is characterised in that described acquisition user property Information includes:

Gather the image of the display location region of described content recommendation；

The user as service object is determined from described image；And

Described user is carried out ontoanalysis and group analytic.

Method the most according to claim 3, described determines as service from described image The user of object, including:

Detect the sight line focal position of the pedestrian in described image and described pedestrian；

Judge whether the sight line focal position of described pedestrian is positioned at the display location of content recommendation；With And

If it is, determine that described pedestrian is the described user as service object.

5. according to the arbitrary described method of claim 3-4, it is characterised in that described to described User carries out ontoanalysis and includes:

According to human body, each user in image is divided into multiple subimage；And

Grader and/or recurrence device is used to be analyzed obtaining user property to described subimage Information；

Wherein, described customer attribute information include following at least one: sex, the age, race, Clothing style, cosmetic style, health status, personality and purchasing power；And

Described personality include following at least one: sense of responsibility, degree of being emotionally stable, export-oriented degree, Degree of opening, affinity, popularity, Confidence and lonely degree to new things.

Method the most according to claim 5, it is characterised in that described described user is entered Row group analytic includes described user grouping, including:

Based on the clothing of user multiple in described image, the correlation degree of attitude and relative position Information, uses grader to classify the multiple users in described image according to social relations； And/or

Based on ontoanalysis result, utilize customer attribute information that users multiple in described image are entered Row cluster.

7. according to the arbitrary described method of claim 2-6, it is characterised in that described acquisition ring Border attribute information includes:

Receive current temporal information by network, described temporal information include following at least one: Current date-time, weather condition, festival information, current hot ticket；And/or

Spatial information corresponding with the display location of described content recommendation is received by network, described Spatial information includes the geographic orientation of display location and/or the terrestrial reference of adjacent domain.

8. according to the arbitrary described method of claim 2-7, it is characterised in that described determine time Content element is selected to include: based on described customer attribute information and/or described environment attribute information, profit Candidate's element content is determined with recommended models collection is incompatible,

Wherein, described recommended models set include following at least one:

According to customer attribute information recommended element and/or the first recommended models of situation elements Set, object elements and/or situation elements by the second recommended models set of combine recommendation, according to Environment attribute information recommendation object elements and/or the 3rd recommended models set of situation elements.

Method the most according to claim 8, it is characterised in that

Described first recommended models set includes at least one customer attribute information and object elements The submodel of interest relation between the attribute of attribute and/or situation elements；

Described second recommended models set includes attribute and/or the scene of at least one object elements The submodel of interest relation between attribute of an element；

Described 3rd recommended models set includes at least one environment attribute information and object elements The submodel of interest relation between the attribute of attribute and/or situation elements.

Method the most according to claim 9, it is characterised in that described based on described use Family attribute information and/or described environment attribute information, in utilizing that recommended models collection is incompatible and determining candidate Hold element, including:

Based on interest-degree statistical data, the submodel in described recommended models set is trained, To determine the parameter of described submodel；

Global energy function is set up based on described recommended models set；

Described global energy function is carried out global optimization solve, obtain so that global energy function Optimum alternating content element；

Wherein, described interest-degree statistical data includes:

Described customer attribute information is to the attribute of described object elements and/or described situation elements The interest-degree statistical data of attribute；

Interest-degree statistical data between the attribute of different object elements；

Interest-degree statistical data between the attribute of different situation elements；

Interest-degree statistical number between attribute and the attribute of described situation elements of described object elements According to；And

Described environment attribute information is to the attribute of described object elements and/or described situation elements The interest-degree statistical data of attribute.

11. methods according to claim 10, it is characterised in that described global energy letter Number includes the first energy function, the second energy function and the 3rd energy function；

Described first energy function includes: based on described first recommended models set, uses classification The attribute of the object elements corresponding with described customer attribute information that device and/or recurrence device calculate And/or the recommendation probability of the attribute of situation elements；

Described second energy function includes: based on described second recommended models set, uses classification The attribute of the described object elements that device and/or recurrence device calculate and/or the attribute quilt of situation elements The probability of combine recommendation；

Described 3rd energy function includes: based on described 3rd recommended models set, uses classification The attribute of the object elements corresponding with described environment attribute information that device and/or recurrence device calculate And/or the recommendation probability of the attribute of situation elements.

12. methods according to claim 10, it is characterised in that described interest-degree is added up Data in the following way at least one obtain: experience set, questionnaire statistics and The data statistics of online shopping site.

13. according to the arbitrary described method of claim 2-12, it is characterised in that described basis Described candidate target element and alternate scenes element synthesis content recommendation, including:

Obtain the placement index of described alternate scenes element, direction index and movement locus index；

Index described according to described index, described direction index and the described movement locus placed Candidate target element merges with described alternate scenes element, to generate Candidate Recommendation content；And

Based on the degree of association between described alternate scenes attribute of an element, described candidate target element Degree of association between attribute and described alternate scenes attribute of an element and candidate target attribute of an element Between degree of association described Candidate Recommendation content is screened, will through screening Candidate Recommendation in Hold and merge, to generate content recommendation.

14. according to the arbitrary described method of claim 2-13, it is characterised in that described method Also include:

The mode that the time-division presents or space division presents is used to show described content recommendation.

15. methods according to claim 14, it is characterised in that described employing space division in Existing mode shows described content recommendation, including:

The display location of described content recommendation is divided into and organizes, with user/user, the son that quantity is equal Region；And

Would correspond to each/group user content recommendation be shown to this user/this group user's sight line In the subregion at place, focal position.

16. methods according to claim 14, it is characterised in that described employing space division in Existing mode shows described content recommendation, also includes:

Follow the tracks of the sight line focal position of described user；And

The change of the sight line focal position according to described user, adjust described each/group user The display location of content recommendation.

17. methods according to claim 14, it is characterised in that the described employing time-division in Existing mode shows described content recommendation, including:

The display location of content recommendation switches at a certain time interval corresponding to each/ The content recommendation of group user.

18. methods according to claim 17, it is characterised in that the described employing time-division in Existing mode shows described content recommendation, also includes:

Follow the tracks of the sight line focal position of described user；And

The change of the sight line focal position according to described user, adjust corresponding to described each/ The displaying angle of the content recommendation of group user.

19. according to the arbitrary described method of claim 9-18, it is characterised in that

Described object elements includes that commodity, described situation elements include ad elements；

The attribute of described object elements include following at least one: the classification of commodity, price, face Normal complexion brand；

The attribute of described situation elements include following at least one: visual style, plot, Be suitable for commodity, personage, time, place and dub in background music.

20. 1 kinds of content recommendation devices, it is characterised in that described device includes:

Acquiring unit, is configured to obtain customer attribute information and/or environment attribute information；And

Synthesis unit, is configured to believe based on described customer attribute information and/or described environment attribute Breath synthesis content recommendation.

21. 1 kinds of content recommendation systems, it is characterised in that described system includes processor and shows Show equipment；

Described display device is configured to show content recommendation；

Described processor includes content recommendation device as claimed in claim 20.