The content of the invention
(1) technical problems to be solved
The purpose of the present invention is a kind of matching method with the dress ornament based on individual subscriber wardrobe, meets the fashion of user
Taste demand.
(2) technical scheme
The present invention is achieved by the following technical solutions:A kind of matching method of individual subscriber dress ornament, including following step
Suddenly:
S1 user takes pictures to the dress ornament in its wardrobe, and imports computer and build the personal wardrobe mould of vision data driving
Type, user shoot dress ornament photo in its wardrobe, so as to use the method for visual analysis structure one " personal wardrobe model ", and produce
The visual data characteristics of one quantization, to represent the concept and method of the personal fashion style of user.
S2 people's wardrobe model is based on network big data application machine learning method and understands faddish trend, there is provided clothing matching
It is recommended that personal wardrobe model is learnt to understand faddish trend using the public data of social networks by machine learning module.
S3 user selects dressing with reference to clothing matching suggestion.
Wherein, in the step S2, training data understanding fashion can be obtained by periodic harvest social network information and is become
Gesture.
Wherein, the step S2 specifically includes following steps:
The clothing matching picture that S21 people's wardrobe model passes through machine learning module automatic mining social media;
S22 calculates certain style clothing matching by the statistical language model method of application natural language processing field
Probability of occurrence order is distributed, to analyze fashion clothing matching;
After user specifies some dress ornament, probability of occurrence of the personal wardrobe model based on clothing matching provides dressing and built S23
View.
Preferably, the step S2 is further comprising the steps of:
The star liked according to user wears the clothes style, the probability of occurrence distributed constant of clothing matching in set-up procedure S22.
Wherein, the step 23 specifically includes following steps:
Some dress ornament specified for user, using a N-gram model, for predicting the dress ornament of next appearance;
The N-gram models are Markov model, i.e. (N-1)-order;
In fashion world, N represents the dress ornament quantity being related to during dress ornament is predicted.
Further, it is further comprising the steps of:
Methods of the S4 by setting up randomness judgment models, to judge the prediction effect of personal wardrobe model.
Wherein, the method for the horizontal scoring of fashion that whole wardrobe is provided also is included in the step S2, specific method is as follows:
The fashion score of whole personal wardrobe model is calculated using the formula P (w1 ..., wm) of statistical language model
Evaluation;Wherein, wm represents a dress ornament, and P represents the fashion degree of various clothing matchings.
Further, the personal wardrobe model can recommended user's dress ornament purchase link, for in its people's wardrobe model
Clothing matching, lift the fashion index of the personal wardrobe model.
Wherein, the step S1 specifically includes following steps:
The dress ornament that S11 goes to take pictures to user using an automatic categorizer is classified, and stamps semantic label;
S12 extracts the multidimensional vision characteristic vector of taken pictures dress ornament using SVMs module;
S13 establishes the personal wardrobe model of a similar statistical language model by visual feature vector.
(3) beneficial effect
Compared with prior art and product, the present invention has the following advantages:
The present invention provides a kind of matching method of personal dress ornament, and the magnanimity fashion on network is analyzed by machine learning method
Picture, faddish trend is understood based on network big data, and " fashion knowledge " that application is got recommends to be based on it for user
The clothing matching of people's wardrobe, quickly meet customization requirement and the fashion demand of client.
Embodiment
Understand for the ease of those of ordinary skill in the art and implement the present invention, below in conjunction with the accompanying drawings and embodiment
The present invention is described in further detail.
As shown in figure 1, the present embodiment provides a kind of matching method of individual subscriber dress ornament, comprise the following steps:
S1 user takes pictures to the dress ornament in its wardrobe, and imports computer and build the personal wardrobe mould of vision data driving
Type;
Wherein, user shoots dress ornament photo in its wardrobe, so as to build " personal wardrobe a mould with the method for visual analysis
Type ", and the visual data characteristics of a quantization are produced, to represent the concept and method of the personal fashion style of user.
Step S1 specifically includes following steps:
The dress ornament that S11 goes to take pictures to user using an automatic categorizer is classified automatically, and stamps semantic label.
In order to obtain the personal wardrobe model of user, we can use certain calibrating method.For example we can remind user
Go to take pictures to the clothes in her wardrobe.In this implementation process, we can go to clap user with an automatic grader
According to dress ornament classify (such as skirt, trousers, footwear etc.) automatically, and stamp semantic label.
S12 extracts the multidimensional vision characteristic vector of taken pictures dress ornament using SVMs module;Using nature language
Outside speech processing (Natural Language Processing, NLP) art processes, there are other schemes, one of scheme
It is exactly the technology learnt using other machines, if SVMs (Support Vector Machines, SVM) is to dress ornament figure
Piece is identified.
In terms of data structure, the model of a fashion dress ornament removes table with the multidimensional characteristic vectors based on object visual signature
Show, and it is what to go to describe this object with semantic label.Wherein it is significant to note that, we use visual signature, without
It is Textuality description, because Textuality description can lose important information, and the visual signature of computer extraction can be more
The substantive characteristics of an object is obtained well and can be used quantifies data-driven.
S13 establishes the personal wardrobe model of a similar statistical language model by visual feature vector.
Wherein, next visual feature vector can be established a similar statistical language model with let us
The fashion model of (Statistical Language Model, SLM), and built with it to be provided for the fashion clothing matching of user
View, or calculate its fashion index.
S2 people's wardrobe model is based on network big data application machine learning method and understands faddish trend, there is provided clothing matching
It is recommended that.
Specifically, when personal wardrobe model is learnt to understand using the public data of social networks by machine learning module
Still trend, step S2 include step in detail below:
The clothing matching picture that S21 people's wardrobe model passes through machine learning module automatic mining social media.
S22 calculates setting style clothing matching by the statistical language model method of application natural language processing field
Probability of occurrence order is distributed, to analyze fashion clothing matching.
This novel implementation will apply algorithm and technology from natural language processing NLP fields.We see
Observe, to solve how to allow different dress ornaments to be collocated with each other, how this problem in natural language processing with find phase
The word mutually coordinated is similar.For example, why " strong team " are than " powerful tea " can more suitably be taken
Match somebody with somebody, although they are correct from English GrammarThis is just as why the collocation of white tee shirt and blue jeans is than purple
The collocation of yellow T-shirt and blue jeans is more preferableIn order to answer this problem, we borrow language material linguistics (corpus
Linguistics) inner a concept, phrase collocation (collocations), the order occurred equivalent to word and combination.Just
As if we analyze substantial amounts of ordinary language text, it finds that " strong tea " are than " powerful tea " have
The higher frequency of occurrences.In our implementation method, we utilize this concept and similar mathematical method and technology hand
They, are applied to this field of fashion clothing matching by section.
For S23 after user specifies some dress ornament, probability of the personal wardrobe model based on clothing matching provides dressing suggesting.
Our purpose is to establish a forecast model, and this model can go conjecture to obtain after some dress ornament is specified
Other suitable dress ornaments, get up with former clothing matching as a kind of style of fashion.We can use natural language processing
A concept in this field, statistical language model (Statistical Language Model, SLM), it represents word
The probability distribution of appearance order.In fashion dress ornament field, each word just represents a dress ornament with certain style.
In order to reach the purpose of prediction, following steps are specifically included:
Some dress ornament specified for user, using a N-gram model, for predicting the dress ornament of next appearance;
The N-gram models are Markov model, i.e. (N-1)-order;
In fashion world, N represents the dress ornament quantity being related to during dress ornament is predicted.
For example, if we need to go to predict a T-shirt based on a shorts and a pair of shoes, then N is equal to 3.
In order to which the concept in these NLP of application and information theory is to fashion world, we also need to use a markov
The characteristic of model:The probability of occurrence (what dress ornament of arranging in pairs or groups) of one future event, some limited have occurred before being only dependent upon
Event (what dress ornament user has selected).
Wherein, it can also be some dress ornament in the present embodiment, and be finally whole wardrobe, there is provided one represents fashion level
Evaluation score.In order to obtain some clothing matching others dress ornament whether fashion, we borrow statistical language model
The concept of (Statistical Language Model, SLM), and the formula P (w1 ..., wn) of statistical language model is used,
(in NLP fields, wn represents some word, and P (w1 ..., wn) represents the probability that some word combinations occur) calculates some clothes
The fashion degree of other dress ornaments of decorations collocation.Wherein, with the present embodiment, each word wm represents a dress ornament, and P represents each
The fashion degree of kind clothing matching.
Removing to assess the probability of some clothing matching, (the fashion picture that we can be gone on mining analysis network is to build this people
Wardrobe model, just look like NLP the same using the inner text that can go to excavate on wikipedia).As an example it is assumed that w1 is certain
The cap of style, w2 are the jackets of certain style, and w3 is the trousers of certain style, then P (w1, w2, w3) just represents these three
The fashion degree that clothing matching gets up.We can also go to calculate with the mode of conditional probability, give the jacket w2 of certain style,
With the trousers w3 of certain style, then the cap w1 of what style and they arrange in pairs or groups it is the most nice, with P (w1 | w2, w3) table
Show.
Further, in order to allow user constantly to lift its fashion after the fashion index of personal wardrobe model is obtained
Index, we can excavate the content (method that social networks is excavated with being described above is similar) of e-commerce website, and provide electricity
Sub- business function.Personal wardrobe model can recommended user's dress ornament purchase link, for the clothing matching in its people's wardrobe, carry
Rise the fashion index of the personal wardrobe model.For example, when this wardrobe model give user recommend one collocation dress ornament, such as
Without this dress ornament, we can provide a purchase link, allow user to leap on e-commerce website and contain fruit user
The page of this dress ornament, user is allowed directly to buy this dress ornament.Personal wardrobe model can recommended user's dress ornament purchase link, use
In with the clothing matching in its people's wardrobe, lift the fashion index of the personal wardrobe model.
The present embodiment, it may also provide a kind of Personalized service.For this wardrobe model of further tuning, and cause
The clothing matching suggestion that this model provides the user with can wear the clothes style closer to the star that this user is liked, Wo Menhui
Provide the user with the model of a upgrade version.The star liked according to user wears the clothes style, clothing matching in set-up procedure S22
Probability of occurrence distributed constant.For example, if the star that a user likes often wears red trousers and matches somebody with somebody white shirt, then this
In the new model of user, it will increase with the probability of the red trousers of white shirt collocation.
In step S2, it can also understand fashion by the method for periodic harvest social network information acquisition training data and become
Gesture, so as to provide clothing matching suggestion.In order to obtain a large amount of training datas, and keep the freshness and trend of these data
Property, it is exactly to excavate the information on social networks that we, which have a critically important means,.The training data is exactly fashion microblogging etc.
The data such as comment number, forwarding number, collection number in social media on clothing matching.We can allow computer to pass through social network
Information learning on network to some faddish trends, such as, periodically go to collect some comments on clothing matching fashion microblogging
Number, forwarding number, collection number etc..If these data are in fast lifting, then they may represent a kind of new trend, if this
A little data are constant for a long time, then if this may represent a stable trend work as in computer discovery social media certain it is bright
Certain fashion wearing of star is got growing concern for, then computer may can take this new fashion wearing
With being judged as a kind of new fashion, and suitable user is recommended in similar wearing collocation.
S3 user selects dressing with reference to clothing matching suggestion.
Methods of the S4 by setting up randomness judgment models, to judge the prediction effect of personal wardrobe model.
In order to assess this fashion model, we can also further utilize NLP concept, such as randomness
(Perplexity), this is to assess the standard method of language model in information theory, for the prediction effect of assessment models.One
The randomness of individual language model is smaller, and model is better.
The matching method for the individual subscriber dress ornament that the present embodiment provides, by a machine learning method, with automatically
Crawl and picture of the study from social media, so as to learn to faddish trend;Then further faddish trend application is got home
Front yard wardrobe, and finally recommend fashion dress ornament to user.In a word, we teach computer voguish style, and allow computer help
User selected from oneself wardrobe suitable dress ornament and arrange in pairs or groups as star fashion.Do not closed in user's wardrobe sometimes
During suitable dress ornament, we also can recommended user buy this dress ornament.Finally, we can go to analyze user's wardrobe pattern and fashion tide
Difference between stream, so as to realize that the method with quantification obtains " the fashion index " of user, to reflect the voguish style of user
It is horizontal.
Above example is only one embodiment of the present invention, its describe it is more specific and in detail, but can not therefore and
It is interpreted as the limitation to the scope of the claims of the present invention.Its concrete structure and size can be adjusted correspondingly according to being actually needed.Should
When, it is noted that for the person of ordinary skill of the art, without departing from the inventive concept of the premise, can also make
Several modifications and improvements, these belong to protection scope of the present invention.