CN111695040A - A fashion product recommendation method, system and device based on emotion tags - Google Patents

Abstract

The invention discloses a fashion product recommendation method, system and device based on emotional tags. The method mainly recommends to users according to the popularity of fashion products and the total weight score of fashion products that conform to the current fashion style and satisfy their own preferences. of fashion items. The method includes: acquiring user's rating information and emotion tag information; establishing a user emotion dictionary according to the user's emotion tag information; acquiring fashion product information, and calculating a popularity score according to the fashion product life cycle; recommending fashion products based on similar users' fashion product recommendation scores Fashion products; the present invention not only considers the objective feedback of users' ratings on fashion products, but also supplements the user's feedback through emotional tags of fashion products, which can more accurately grasp the user's preference and improve the performance of recommendation; by predicting the life of fashion products Cycle, according to the change of the popularity of fashion products over time, it can recommend to users fashion products that meet both the user's own preference and the current fashion style.

Description

A fashion product recommendation method, system and device based on emotion tags

technical field

The invention relates to the field of recommendation systems in computer technology, in particular to a method, system and device for recommending fashion products based on emotion tags.

Background technique

In the process of fashion product recommendation, the most important consideration is how to recommend fashion products that meet both the current fashion style and the user's own preferences. Fashion products have different life cycles. Once a fashion product is introduced to the market, its popularity will change over time, and after the life cycle of a fashion product, it will eventually be replaced by a new fashion product. The emotional tags added to fashion products reflect the user's emotional preference for fashion products to a certain extent. Therefore, the fashion product recommendation method based on emotional tags is to recommend fashion products that conform to the current fashion style and satisfy their own preferences according to the popularity of fashion products and the score of the total weight of fashion products.

SUMMARY OF THE INVENTION

In order to solve the above problems, the present invention provides a fashion product recommendation method, system and device based on emotion tags, which recommends corresponding fashion items according to the life cycle of the fashion items and the user's own preferences reflected by the emotion tags.

In order to achieve the above purpose, the technical solution adopted in the present invention is: a method for recommending fashion products based on emotional tags, the method comprising the following steps:

S1, for the fashion products that the user has purchased, obtain the user's rating information for the fashion products and the emotional score of the added tags; for the fashion products that the user has browsed, obtain the emotional scores of the added tags;

S2, establish the user's emotional dictionary;

S3, obtain the information of the fashion product, the fashion product information at least includes the predicted life cycle of the fashion product, and calculate the popularity score of the fashion product according to the life cycle of the fashion product;

S4, according to the S1 user's rating information for fashion products and the emotional score of the label, calculate the total weight score of fashion products and sort them, and obtain a ranking list of the total weight score of fashion products; the calculation process of the total weight score of fashion products is as follows:

Calculate the weighted score of fashion products based on the scoring information;

Calculate sentiment-based fashion weighting scores;

The total weighted score of fashion products=(1-α)*the total weighted score of fashion products based on rating information+α*the weighted score of fashion products based on emotion;

S5, based on the total weight score of each user's fashion products obtained in S4, using the total weight score of each user u to calculate the average total weight score μ _u of each user, and then using the similarity principle to calculate the Pearson correlation coefficient, Get a sorted list of similar users v of user u;

S6, according to the user's fashion product weight score result obtained in S4 and the fashion product popularity score obtained in S3, calculate the fashion product recommendation score and sort it, and obtain a recommendation score ranking list;

Fashion product recommendation score = popular score * total weight score of fashion products

According to the sorted list of similar users obtained in step 5, fashion items are recommended to similar users from the list of fashion item recommendation scores.

In S2, the user's emotional dictionary records the emotional tags and emotional scores added by the user, and the user can add and delete emotional tags at any time, and modify the emotional scores.

In S3, the information of the fashion item further includes release date, brand and price.

In S3, the popularity score of fashion items:

Among them: T is the predicted life cycle of fashion products, t is the number of days after the release of fashion products, as the number of days of release increases, the popularity score of fashion products is getting lower and lower, and the minimum value of F _Score is 0. For classic fashion products , and its popularity score is always 1.

In S4, the weighted score calculation of fashion products based on the scoring information:

where r _u,i is the user's rating value for fashion items, r _u,i (h) the normalized rating value of the label h added by the user for fashion item i, vectorize each user's rating, and then normalize it is a unit vector:

I is the total number of all fashion items rated by users.

In S4, the sentiment-based fashion weight score is calculated as follows:

(1) Delete the special characters contained in the label;

(2) Remove proper nouns in labels;

(3) Calculate the sentiment score of the label, use the user's sentiment dictionary to calculate the sentiment score of the label, if the label exists in the user's sentiment dictionary, use the score in the sentiment dictionary as the label's sentiment score:

w _e (hu _,i )=EmotionScore(hu _,i )

Where EmotionScore(h _u,i ) is the sentiment value of the label hu, _i in the sentiment dictionary, if the sentiment dictionary contains the label, the sentiment score of the label is the sentiment score in the sentiment dictionary;

If multiple tags exist in the user's sentiment dictionary at the same time, the sentiment score is calculated according to the sentiment values of multiple tags:

where set is a set consisting of multiple tags, |set| is the number of all tags in the set, and |set _emotion | is the number of emotion tags in the set.

In S5, use the total weight score of each user u to calculate the average total weight score μ _u of each user:

I _u represents a set of serial numbers of fashion items that have been evaluated by user u; the Pearson correlation coefficient can be used to measure the similarity of the rating vector between user u and user v Sim(u, v), the difference between user (row) u and v The Pearson correlation coefficient between is defined as follows:

where I _u ∩ I _v is the set of total tag weight scores known to both user u and user v.

A fashion product recommendation system, comprising: an information acquisition module, used for obtaining the user's rating information for the fashion product and the emotional score value of the added tag, obtaining the user's emotional dictionary and obtaining the information of the fashion product;

The data processing module is calculated based on the user's rating information for the fashion item and the sentiment score added to the tag, and obtains the user's sentiment dictionary and information about the fashion item: the fashion item weight score based on the rating information, and the emotion-based fashion item weight score , the total weight score of fashion products, the popularity score of fashion products, the fashion product recommendation score and the list of fashion product recommendation scores; based on the user's total weight score of fashion products, similar users are calculated using the similarity principle;

The recommendation module recommends fashion items to similar users based on the fashion item recommendation score, the fashion item recommendation score list and similar users.

A fashion product recommendation device, comprising a processor, a memory, an information acquisition device and a recommendation result output device, the processor and the memory are connected through an I/O interface, the information acquisition device is connected to the input end of the processor, and the recommendation result output device is connected to the processor. The output end; the memory stores an executable computer program, and the processor can execute the fashion product recommendation method of the present invention when the executable computer program is executed, and display the recommendation result through the recommendation result output device.

Compared with the prior art, the present invention has at least the following beneficial effects: the present invention not only considers the objective feedback of the user's rating on the fashion product, but also supplements the user's feedback through the emotional label of the fashion product, and can more accurately grasp the user's preference. , to improve the performance of recommendation; by predicting the life cycle of fashion products, according to the changes in the popularity of fashion products over time, it can recommend to users fashion products that meet both the user's own preference and the current popular style; the use of emotion dictionary can be very convenient to manage And maintain the emotional tag added by the user, which can quickly adjust the recommended fashion products according to the user's emotional preference changes.

Description of drawings

Figure 1 is a flow chart of fashion product recommendation based on emotion tags.

Detailed ways

Embodiments of the present invention will be described below with reference to FIG. 1 .

The fashion product recommendation method based on emotion tags has the following specific steps:

Step 1: For the fashion items that the user has purchased, obtain the user's rating information for the fashion item and the sentiment score of the added tag; for the fashion item that the user has browsed, obtain the sentiment score of the added tag; the range of the sentiment score From -1 to 1, a negative sentiment score indicates that the user has a negative sentiment. The closer it is to -1, the deeper the negative sentiment; a positive sentiment value indicates that the user has a positive sentiment, and the closer it is to 1, the The higher the level of positive emotion.

Step 2, establishing an emotion dictionary of the user; recording the emotion tags and emotion scores added by the user, and adding and deleting emotion tags and modifying emotion scores at any time.

Step 3: Obtain the information of the fashion product, the fashion product information includes the predicted life cycle, brand and price of the fashion product; if the monthly fashion product is issued, the life cycle of the fashion product is predicted to be 30 days; if the quarterly fashion product is released , the life cycle of the fashion product is predicted to be 90 days; if the annual fashion product is released, the life cycle of the fashion product is predicted to be 365 days.

According to the life cycle of fashion products, the popularity score of fashion products is calculated:

Among them: T is the predicted life cycle of fashion products, and t is the number of days after the fashion products are released. With the increase of release days, the popularity score of fashion products is getting lower and lower, and the minimum value of F _Score is 0; for classic fashion products, its popularity score is always 1.

Step 4, according to the user's rating information of the fashion product in step 1 and the emotional score of the added tag, calculate the total weight score of the fashion product and sort it;

The calculation steps and methods of the total weight score of fashion products are as follows:

1. Calculation of fashion product weight score based on scoring information:

where r _u,i is the user's rating value for fashion items. r _u,i (h) The normalized rating value of the label h added by the user for fashion item i. If raw ratings are used as label weights, there may be a bias, as the range of ratings given to fashion items varies from user to user. So vectorize each user's rating, then normalize it to a unit vector:

2. Emotion-based fashion weight score calculation:

Sentiment-based weighted scores for fashion items, represented by the sentiment value of the tag, to obtain the sentiment score, perform the following steps for each tag:

(1) Delete the special characters contained in the label;

(2) Remove proper nouns in labels. Proper nouns do not accurately reflect sentiment, so the proper nouns are removed when calculating sentiment scores;

(3) Calculate the sentiment score of the tag. Use the user's sentiment dictionary to calculate the sentiment score of the label, if the label exists in the user's sentiment dictionary, use the score in the sentiment dictionary as the label's sentiment score:

w _e (hu _,i )=EmotionScore(hu _,i )

Where EmotionScore(hu _,i ) is the sentiment value of the label hu, _i in the sentiment dictionary, if the sentiment dictionary contains this label, the sentiment score of the label is the sentiment score in the sentiment dictionary.

If multiple tags exist in the user's sentiment dictionary at the same time, the sentiment score is calculated according to the sentiment values of multiple tags:

where set is a set consisting of multiple tags, |set| is the number of all tags in the set, and |set _emotion | is the number of emotion tags in the set.

If the label does not exist in the sentiment dictionary, the sentiment value of the label is 0 this time, and the user can add the label to the sentiment dictionary and set the sentiment score for the subsequent sentiment value calculation of the label.

Calculation of the total weight score of fashion products:

weight(h _u,i )=(1-α)*w(h _u,i )+α*w _e (h _u,i )

where α is a parameter that controls the influence of sentiment labels, if the label has no sentiment value, only the rating-based fashion item weight is used to calculate the total weight, and if the fashion item has no rating value, only the sentiment-based fashion item weight is used to calculate the total weight .

Step 5, based on the total weight score of fashion products obtained in step 4, construct an m×n label total weight score matrix R=[w _uj ] of users and fashion products, where w _uj is the user u calculated in step 4 to fashion. The total weight score of fashion products of product j;

Using the total weight score of each user u, calculate the average total weight score μ _u of each user:

I _u represents the serial number set of fashion items that have been evaluated by user (row) u; Pearson correlation coefficient can be used to measure the similarity of rating vectors between user u and user v Sim(u, v), user (row) u The Pearson correlation coefficient between v and v is defined as follows:

where I _u ∩ I _v is the set of total tag weight scores known to both user u and user v;

Step 6: Calculate the popularity scores of the corresponding fashion items on that day, and then calculate and obtain the fashion item recommendation scores, and sort them according to the size of the fashion item recommendation scores to obtain the final fashion item recommendation score list. According to the similar users obtained in step 5, A sorted list of the total weights of fashion items of similar users, and recommend fashion items to other similar users from the list of fashion item recommendation scores;

Fashion recommendation score calculation method:

Rec _Score =F _Score *weight( _hu,i )

The present invention also provides a fashion product recommendation system, comprising: an information acquisition module for obtaining the user's rating information for the fashion product and the emotional score added to the tag, obtaining the user's emotional dictionary and obtaining the information of the fashion product;

The data processing module is calculated based on the user's rating information for the fashion item and the sentiment score added to the tag, and obtains the user's sentiment dictionary and information about the fashion item: the fashion item weight score based on the rating information, and the emotion-based fashion item weight score , the total weight score of fashion products, the popularity score of fashion products, the fashion product recommendation score and the list of fashion product recommendation scores; based on the user's total weight score of fashion products, similar users are calculated using the similarity principle;

The recommendation module recommends fashion items to similar users based on the fashion item recommendation score, the fashion item recommendation score list and similar users.

A fashion product recommendation device, comprising a processor, a memory, an information acquisition device and a recommendation result output device, the processor and the memory are connected through an I/O interface, the information acquisition device is connected to the input end of the processor, and the recommendation result output device is connected to the processor The output end; the memory stores an executable computer program, and the processor can execute the fashion product recommendation method described in any one of claims 1-7 when the processor executes the executable computer program, and the recommendation result is displayed through the recommendation result output device. .

As a preferred embodiment, the information acquisition device and the recommendation result output device both use a touch display, and the touch display is connected to the processor through an I/O interface.

When the fashion product recommendation device executes step 1 by executing the computer program, the information acquisition device obtains the fashion products that the user has purchased, and obtains the user's rating information for the fashion product and the emotional score added to the tag; For fashion products, obtain the sentiment score of the added tag; the sentiment score ranges from -1 to 1, and a negative sentiment score indicates that the user has negative emotions. The closer to -1, the deeper the negative emotion; A positive value means that the user has a positive emotion, the closer it is to 1, the deeper the positive emotion; and stored in the memory,

When the fashion product recommendation device performs step 2 by executing the computer program, the information acquisition device records the emotional tags and emotional scores added by the user, and can add and delete emotional tags at any time. the information is stored in memory;

The memory also stores an instruction set for automatically acquiring information, the processor is connected to the network transmitter, and the processor can execute the instruction set for automatically acquiring information to acquire the information of steps 1 and 2 of the present invention from the network, and the fashion When the product recommendation device executes steps 1 and 2 by executing a computer program, the information may be obtained by the processor executing an instruction set for obtaining information, or the information may be manually input by the user.

When the fashion product recommendation device executes step 3 by executing the computer program, the information of the fashion product is obtained, and the information is stored in the memory by the processor; at the same time, the processor calculates the popularity score of the fashion product according to the life cycle of fashion spelling, Simultaneously send popular scores to memory;

When the fashion product recommendation device performs step 4 by executing the computer program, the processor calculates the fashion product weight score based on the scoring information, the emotion-based fashion product weight score and the total fashion product weight score, and calculates the fashion product weight score, based on The emotional fashion item weighting score and the fashion item total weighting score are sent to the memory;

When the fashion product recommendation device executes step 5 by executing the computer program, based on the fashion product total weight score of each user obtained in step 4, the average total weight score of each user is calculated by using the label total weight score of each user. , and then use the similarity principle to calculate the correlation coefficient to obtain a list of similar users, and send the obtained results to the memory for storage;

When the device for recommending fashion items executes step 6 by executing the computer program, the processor reads from the memory the popularity score of the fashion items obtained in step 3 and the fashion item weight score result of the user obtained in step 4, and recommends the scores according to the fashion items = Popularity score*total weight score of fashion products is calculated to obtain fashion product recommendation scores and sorted; at the same time, read the similar user data obtained in step 5, and recommend fashion products to similar users from the fashion product recommendation score list; also recommend the results through the recommendation result output device exhibit.

Claims (9)

1. A fashion recommendation method based on emotion labels is characterized by comprising the following steps:

s1, acquiring the grading information of the fashion and the emotion score of the added label for the fashion purchased by the user; acquiring the emotion score of the added label for the fashion browsed by the user;

s2, establishing an emotion dictionary of the user;

s3, obtaining information of the fashion, wherein the information of the fashion at least comprises a predicted life cycle of the fashion, and calculating the popularity score of the fashion according to the life cycle of the fashion;

s4, calculating and sequencing the total weight scores of the fashion according to the grading information of the user to the fashion and the emotion scores of the labels at S1 to obtain a sequence list of the total weight scores of the fashion; the fashion total weight score calculation process is as follows:

calculating a fashion weight score based on the scoring information;

calculating a fashion weight score based on emotion;

a total fashion weight score ═ (1- α) — total fashion weight score based on the scoring information + α — + an emotional fashion weight score;

s5, based on the fashion total weight score of each user obtained in S4, the average total weight score mu of each user is calculated by using the label total weight score of each user u_uThen, calculating a Pearson correlation coefficient by utilizing a similarity principle to obtain a ranking list of similar users v of the user u;

s6, calculating and sequencing the recommendation scores of the fashion according to the weight score result of the fashion of the user obtained in the S4 and the popularity score of the fashion obtained in the S3 to obtain a recommendation score sequencing list;

fashion recommendation score (popularity score) fashion total weight score

And recommending the fashion to the similar users from the fashion recommendation score list according to the similar user ranking list obtained in the step 5.

2. The method for recommending fashion goods based on emotion label as recited in claim 1, wherein in S2, the emotion dictionary of the user records emotion labels added by the user and emotion scores, and the emotion labels are added and deleted by the user at any time, and the emotion scores are modified.

3. The emotion tag-based fashion recommendation method of claim 1, wherein in S3, the fashion information further includes a release date, a brand and a price.

4. The emotion tag-based fashion recommendation method of claim 1, wherein in S3, the fashion popularity score:

wherein: t is the predicted life cycle of the fashion product, T is the number of days after the fashion product is issued, the fashion product popularity score is lower along with the increase of the number of issued days, and F_ScoreThe minimum value is 0, and the popularity score of the fashion product of the classic style is always 1.

5. The emotion tag-based fashion recommendation method of claim 1, wherein in S4, the fashion weight score based on the scoring information is calculated by:

wherein r is_u,iIs the value of the user's score for fashion, r_u,i(h) Vectorizing the score of each user for the normalized score value of the label h added by the user for the fashion item i, and then normalizing the score to be a unit vector:

i is the total number of all fashion items scored by the user.

6. The emotion tag-based fashion recommendation method of claim 1, wherein the emotion-based fashion weight score is calculated as follows in S4:

(1) deleting the special characters contained in the label;

(2) removing proper nouns in the tag;

(3) calculating the emotion scores of the tags, calculating the emotion scores of the tags by using an emotion dictionary of the user, and if the tags exist in the emotion dictionary of the user, using the scores in the emotion dictionary as the emotion scores of the tags:

w_e(h_u,i)＝EmotionScore(h_u,i)

wherein EmotionScore (h)_u,i) Is label h in emotion dictionary_u,iIf the label is contained in the emotion dictionary, the emotion score of the label is the emotion score in the emotion dictionary;

if a plurality of tags exist in the emotion dictionary of the user at the same time, calculating emotion scores according to the emotion values of the tags:

where set is a set composed of multiple tags, | set | is the number of all tags in the set, | set |, which is a set composed of multiple tags_emotionAnd | is the number of emotion labels in the set.

7. The method of claim 1, wherein in step S5, the average total weight score μ for each user is calculated using the total weight score of the tags for each user u_u：

I_uA set of numbers representing fashion items that have been evaluated by user u; pearson correlation coefficients may be used to measure the similarity Sim (u, v) of the scoring vector between user u and user v, and are defined as follows:

wherein I_u∩I_vIs a set of label total weight scores known to both user u and user v.

8. A fashion recommendation system, comprising: the information acquisition module is used for acquiring the grading information of the fashion product and the emotion score of the added label from the user, acquiring an emotion dictionary of the user and acquiring the information of the fashion product;

the data processing module is used for calculating and obtaining the following information based on the grading information of the fashion product and the emotion score of the added label of the user, the emotion dictionary of the user and the information of the fashion product: the method comprises the following steps of (1) obtaining a fashion weight score based on scoring information, a fashion weight score based on emotion, a total fashion weight score, a fashion popularity score, a fashion recommendation score and a fashion recommendation score list; calculating to obtain similar users by utilizing a similarity principle based on the total weight score of the fashion products by the users;

and the recommending module is used for recommending the fashion products to the similar users based on the fashion product recommending scores, the fashion product recommending score list and the similar users.

9. The fashion recommendation device is characterized by comprising a processor, a memory, an information acquisition device and a recommendation result output device, wherein the processor is connected with the memory through an I/O interface; the memory stores an executable computer program that, when executed, enables the processor to perform the fashion recommendation method of any one of claims 1-7 and present the recommendation via a recommendation output device.

Images