CN110634051A - Fresh agricultural product recommendation method based on multi-granularity fuzzy data - Google Patents

Fresh agricultural product recommendation method based on multi-granularity fuzzy data Download PDF

Info

Publication number
CN110634051A
CN110634051A CN201910865268.4A CN201910865268A CN110634051A CN 110634051 A CN110634051 A CN 110634051A CN 201910865268 A CN201910865268 A CN 201910865268A CN 110634051 A CN110634051 A CN 110634051A
Authority
CN
China
Prior art keywords
granularity
user
product
recommendation
fresh agricultural
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910865268.4A
Other languages
Chinese (zh)
Other versions
CN110634051B (en
Inventor
陈秀明
彭明星
杨颖�
曹红兵
南淑萍
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fuyang Normal University
Original Assignee
Fuyang Normal University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fuyang Normal University filed Critical Fuyang Normal University
Priority to CN201910865268.4A priority Critical patent/CN110634051B/en
Publication of CN110634051A publication Critical patent/CN110634051A/en
Priority to ZA2020/05623A priority patent/ZA202005623B/en
Application granted granted Critical
Publication of CN110634051B publication Critical patent/CN110634051B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0282—Rating or review of business operators or products
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0603—Catalogue ordering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/02—Agriculture; Fishing; Mining

Abstract

The invention discloses a fresh agricultural product recommendation method based on multi-granularity fuzzy data, which comprises the following steps: acquiring characteristic parameters of a user and a product; constructing a feature model based on the feature parameters, thereby determining user preference parameters; and inputting the user preference parameters into a multi-granularity fuzzy data recommendation model to obtain a recommendation result, and feeding back the recommendation result to the user. The scheme provided by the invention not only takes the similarity between the user preference product and the recommended product into consideration, but also considers the user satisfaction. Therefore, the product which accords with the preference behavior of the user can be recommended in the recommendation, and the optimal recommendation based on the recommendation can be generated for the user. This will have very big realistic meaning to present fresh agricultural product electricity merchant platform, not only can improve the conversion rate, bring higher income, can also promote user experience, reinforcing user's viscosity to this is outstanding in the market that fresh agricultural product electricity merchant competition is more and more intense at present and is enclosed.

Description

Fresh agricultural product recommendation method based on multi-granularity fuzzy data
Technical Field
The invention relates to the technical field of agricultural product recommendation, in particular to a method for recommending fresh agricultural products based on multi-granularity fuzzy data.
Background
With the rapid development of the internet and the e-commerce industry, people also choose to shop online more often in the selection of shopping modes nowadays. The more and more fresh agricultural products are purchased on line and are well known and used by people, and the method becomes the most popular fresh agricultural product purchasing mode for many young people.
At present, the recommendation technology of each large e-commerce platform system tends to be mature, and recommendation information becomes one of the most convenient information acquisition modes for a large number of internet users. Diversified recommendation services and personalized recommendation technologies aiming at users not only provide a convenient information channel, but also enhance the viscosity of the users. However, in the existing fresh agricultural product e-commerce system, the recommendation effect for fresh agricultural products is not very ideal, most fresh agricultural product e-commerce systems provide more choices for users, and simultaneously generate the problem of 'information overload', which results in a large number of repeated recommendations or recommendation results that are not in line with the expectations of users, and the recommendation method can not hold the consumption will of the users on the fresh agricultural products, wastes a large amount of time and cost, and directly causes that products preferred by the users can not be smoothly recommended to hands. Therefore, how to accurately find the fresh agricultural products which accord with the user preference from the massive fresh agricultural product recommendation list aiming at the characteristics of the fresh agricultural products becomes a problem to be solved urgently at present.
Disclosure of Invention
In order to solve the problems, the invention provides a fresh agricultural product recommendation method based on multi-granularity fuzzy data, which can improve the purchasing power and satisfaction of a user on a fresh agricultural product e-commerce platform.
The invention provides a fresh agricultural product recommendation method based on multi-granularity fuzzy data, which comprises the following steps of:
acquiring characteristic parameters of a user and a product;
constructing a characteristic model according to the characteristic parameters, and determining user preference parameters according to the characteristic model;
and inputting the user preference parameters into a multi-granularity fuzzy data recommendation model to obtain a recommendation result, and feeding back the recommendation result to the user.
Further, the user characteristic parameters include: age, academic history, income, and marital status; the product characteristic parameters comprise: product quality, price and logistics time.
Further, the inputting the user preference parameter into the multi-granularity fuzzy data recommendation model to determine the recommendation result specifically includes:
constructing a multi-granularity fuzzy data processing model by adopting multi-granularity grading proportion;
synthesizing a multi-granularity fuzzy data processing model under multiple modes to obtain a comprehensive multi-granularity effective score;
calculating the similarity between the recommended product and the user preference product according to the satisfaction degree of the user to the recommended product;
and sequencing the recommended products according to the similarity to generate a recommendation result.
Further, the multi-granularity score is proportional, and the expression is as follows:
Figure BDA0002201080050000021
the expression of the comprehensive multi-granularity effective score is as follows:
Figure BDA0002201080050000022
wherein, XatScore for t total data at granularity b, NkmThe multi-granularity score ratio under m different multi-granularity modes is shown, t is the total number of data pieces, and m is the number of different multi-granularity modes.
Further, the satisfaction degree P of the user on the recommended productvThe expression is as follows:
Figure BDA0002201080050000023
wherein N ismaxA comprehensive multi-granularity effective score for the user's preferred ideal product; n is a radical ofkvAnd v is the ratio of the multiple particle size scores of the v product, and v is the v product.
The embodiment of the invention provides a fresh agricultural product recommendation method based on multi-granularity fuzzy data, which has the following beneficial effects compared with the prior art:
the method for recommending the fresh agricultural products based on the multi-granularity fuzzy data, provided by the invention, has the advantages that the fresh agricultural products of fruits, vegetables and aquatic meats are well sorted in the recommendation. Most importantly, the recommendation ranking is not only based on the similarity of the user preferred product and the recommended product, but also takes the user satisfaction into account. Therefore, the product which accords with the preference behavior of the user can be recommended in the recommendation, and the optimal recommendation based on the recommendation can be generated for the user. This will have very big realistic meaning to present fresh agricultural product electricity merchant platform, not only can improve the conversion rate, bring higher income, can also promote user experience, reinforcing user's viscosity to this is outstanding in the market that fresh agricultural product electricity merchant competition is more and more intense at present and is enclosed.
Drawings
FIG. 1 is a schematic flow chart of a method for recommending fresh agricultural products based on multi-granularity fuzzy data according to the present invention;
FIG. 2 is a proposed model operation mechanism provided by the present invention;
FIG. 3 is a graph of satisfaction ranking trends provided by the present invention;
FIG. 4 is a graph of correlation coefficient ranking trends provided by the present invention;
fig. 5 is a graph of the satisfaction and the mean correlation coefficient contrast trend provided by the present invention.
Detailed Description
An embodiment of the present invention will be described in detail below with reference to the accompanying drawings, but it should be understood that the scope of the present invention is not limited to the embodiment.
Referring to 1-2, the invention provides a method for recommending fresh agricultural products based on multi-granularity fuzzy data, which comprises the following steps:
step S1, acquiring characteristic parameters of users and products;
step S2, constructing a characteristic model according to the characteristic parameters, and determining user preference parameters according to the characteristic model;
and step S3, inputting the user preference parameters into the multi-granularity fuzzy data recommendation model to obtain a recommendation result, and feeding the recommendation result back to the user.
The specific analysis for the above step S1 is as follows:
user analysis: compared with the daily purchase of fresh agricultural products in vegetable markets, farmer markets, supermarkets and the like, the online purchase of fresh agricultural products is completely different. The online shopping user group directly determines the development of the fresh agricultural product platform and even the industry, and the individual characteristic difference of the users determines different attitudes of online shopping of the fresh agricultural products. Different user groups, such as age, academic history, income, marital status, etc., will have different consumption preferences.
Product analysis: when purchasing fresh agricultural products on the internet, consumers can perceive greater consumption risk due to the different characteristics of the fresh agricultural products from other products. Several product characteristics that are critical to the user are product quality, price, logistics, etc., which all form a prospective in the user's mind. Only when the product reaches the psychological expectation of the user will the user purchase it. If several aspects do not reach the user's expectation, it means that the user will give up purchasing the fresh agricultural product on line. Therefore, the recommendation algorithm of the E-commerce platform for fresh agricultural products can enhance the purchase expectation of the user only when the product characteristics are implemented, and the recommendation service is successfully completed.
In addition, compared with other products for online shopping, the fresh agricultural products have very different inherent characteristics, so that the characteristics of the fresh agricultural products have certain interference in the online transaction process of the fresh agricultural products. When the fresh agricultural product recommendation model is constructed, the main characteristics of the on-line fresh agricultural products need to be deeply analyzed, so that errors can be avoided in the model construction of the recommendation algorithm. The online transaction of the fresh agricultural products is summarized as follows:
(1) the variety richness of the same fresh agricultural products is not high. Compared with other online shopping products, the same fresh agricultural products are not rich in difference, and the fresh agricultural products which are prepared by the preference of users are often unique. In the case of apple, there are not many options available to a user for online purchase of apples, and brands such as red fuji, national light have essentially taken up most of the user's preferences.
(2) The geographical, seasonal and climatic effects are significant. The production of a large portion of fresh agricultural products is influenced by several factors. For example, lettuce and spinach are both mature in spring, and cabbage, cauliflower and the like are more in winter. Similarly, the growth and variety of fresh agricultural products are very different according to different geographical positions and climates.
(3) And (4) considering nutrition collocation factors. Nowadays, consumers increasingly pay attention to nutrition supplement and intake, and dieticians continuously emphasize that the nutrition is balanced in matching with the diet. Therefore, when the user buys fresh agricultural products, more and more nutrient elements behind the products are considered to be balanced and purchased. For example, when purchasing meat, the user also typically purchases vegetables and fruits to ensure balanced intake of proteins and vitamins. Therefore, the nutrition collocation factors of the fresh agricultural products have important influence on the preference of users.
The specific analysis of the above steps S2 to S3 is as follows:
recommendation method service mode
The recommended methods are now widely used. Unlike the initial stage of recommending system application, even though the user size is huge, the user has very definite requirements, so that the user preference features are well extracted, and the user model is well established. But the recommended systems have developed to the present day, the complexity of products has far exceeded the initial stages, and the data volume of products and users has risen to a very large volume. Based on such background, in the service mode of the recommendation method, not only the characteristic data of the user and the product needs to be well processed, but also the feedback needs to be accurately given to the platform, so that the continuous reinforcement can be supplemented.
The service mode of the fresh agricultural product recommendation method has the advantages that the user and the fresh agricultural product e-commerce platform are organically combined, the user is guaranteed to obtain the best recommendation service, and the platform is guaranteed to continuously know the user in feedback. The service modes of the recommended methods have been applied to practice and can be basically classified into two categories: the first category is user-dominated and the second category is product-dominated. The first category focuses on constructing user preferences and providing recommendation services according to the user preference characteristics. The second category is mainly to refine product features, classify the product features in advance according to the product features, and finally match the product features with users to provide recommendation services. The invention adopts the first type of recommendation method service mode which takes user preference as the leading factor to research, applies the recommendation method service mode to the fresh agricultural product e-commerce platform, and constructs a recommendation model to recommend satisfied fresh agricultural products for users.
According to the online trading characteristics of the fresh agricultural products, a fresh agricultural product recommendation model is constructed by utilizing multi-granularity fuzzy data and applying a product attribute-based collaborative filtering algorithm. The research object of the invention is the fresh agricultural products, the similarity of the fresh agricultural products is firstly calculated according to the recommendation algorithm based on the fresh agricultural products, and then the recommendation list is generated for the user according to the similarity of the fresh agricultural products and the user preference.
Recommendation mechanism
In the multi-granularity fuzzy data recommendation model for fresh agricultural products, the recommendation mechanism of the model is that the user preference and the attributes of the fresh agricultural products are respectively extracted to obtain characteristic attributes through the analysis of multi-granularity fuzzy data, products with corresponding attributes are selected to be matched according to the user preference, and finally, satisfactory fresh agricultural products are recommended for the user through recommendation service. The recommendation mechanism of the multi-granularity fuzzy data recommendation model for fresh agricultural products can be expressed as shown in fig. 2:
referring to fig. 2 in particular, a recommendation mechanism of the multi-granularity fuzzy data recommendation model for fresh agricultural products can be seen from fig. 2. Firstly, characteristic analysis is needed to be carried out on users and products; secondly, extracting feature data of the user and the product to construct a feature model, and determining user preference. And finally, processing the fuzzy data of the fresh agricultural product platform by using a multi-granularity idea, establishing a multi-granularity fuzzy data recommendation model of the fresh agricultural product, and generating recommendation for the user by combining the user preference and the product similarity. And meanwhile, the result is fed back to the platform, so that the platform can recommend and adjust different types of user feedback data.
Recommendation model
The recommendation model uses a collaborative filtering algorithm, wherein the most important step is to calculate the similarity through multi-granularity fuzzy data of fresh agricultural products. When processing multi-granularity fuzzy data, the multi-granularity fuzzy data needs to be validated, and a validity score N is adoptedjMeaning that the effective score will be weighted down for scores at multiple granularities, making NjThe method has more reference and calculation values in the final similarity calculation, so that the recommendation model is more accurate. Multi-granularity score ratio:
Figure BDA0002201080050000061
x in the formulaatThe score is t total data under granularity b, and t is the total data number. N is a radical ofkThe score fraction at this granularity is indicated. In the multi-granularity fuzzy data, the reference value of putting the scores of a single multi-granularity mode into the whole is limited, so that the scores in the single multi-granularity mode can be put into a plurality of multi-granularity modes to comprehensively calculate the fuzzy data. And (3) comprehensive multi-granularity effective scoring:
the average score ratio in the m multi-granularity modes can be calculated by accumulating the score ratios of the products in the m different multi-granularity modes and dividing the accumulated score ratios by m, and the average score ratio can be used as an effective score to be applied to the calculation of subsequent similarity. Because a fresh agricultural product, whether a single granularity score or a multi-granularity score of only one mode cannot be referenced, the application of the comprehensive multi-granularity effective score to the fresh agricultural product recommendation algorithm is the most accurate and available method, and the recommendation result obtained by the method is also the most accurate and most in line with the desire of the consumer.
The key of the recommendation system is to calculate the similarity of the fresh agricultural products according to the multi-granularity fuzzy data. At present, three calculation methods commonly used in the collaborative filtering algorithm for calculating the similarity are Euclidean distance, Pearson correlation coefficient and cosine similarity respectively. The three similarity calculation methods are suitable for different environments, and the respective advantages and disadvantages are different. The Euclidean distance calculates the similarity through the distance between two points, the reflected similarity is too single, and the Euclidean distance is not suitable for calculating the similarity of fresh agricultural products. Therefore, the other two methods are adopted to construct the recommendation model, and the recommendation model is selected according to the recommendation result after the comparison and comprehensive consideration.
Firstly, calculating similarity by using a cosine similarity calculation method:
Figure BDA0002201080050000072
where u represents a product that the user purchased or produced a behavior, and v represents a similar product to be recommended. And calculating the similarity of the two types of data by using the multi-granularity fuzzy data processed by the two types of data through a formula.
Figure BDA0002201080050000073
The greater the similarity between the consumer fresh agricultural product u and the fresh agricultural product v, the higher the probability of success in recommending the fresh agricultural product to the consumer.
In practical application, the similarity recommendation algorithm needs to be optimized properly. The recommendation result obtained by pure similarity calculation can be close to the product consumed by the user, but when the user does not consume a certain product or consumes a small amount of products, the similarity recommendation can face the cold start problem without a large amount of user behavior basis, even when the user consumes a certain low-quality product, the recommendation result can be changed into a set of low-quality products, and the recommendation is wrong.
To solve the cold start and wrong recommendation problems, a satisfaction is introduced to optimize this approach. The user preferred products are idealized here, and it is assumed that the user preferred products are the best of this category, which also corresponds exactly to the consumer psychology of the public. Therefore, a recommendation similar to the best product can be directly generated for the user in the cold start state, and the wrong recommendation can be directly reduced to the end in the sorting due to the fact that the difference between the low-quality product and the best product is too large. Satisfaction degree:
Figure BDA0002201080050000081
where P represents the user's satisfaction with the recommended product. N is a radical ofmaxThe composite multi-granularity effective score of an ideal product representing the user's preference is 100% in the ideal state, i.e., 1. Namely, the similarity formula is optimized to be the similarity between the recommended product and the ideal product preferred by the user, so as to express the satisfaction degree of the user on the recommended product. Therefore, the optimized satisfaction formula not only solves the problems of cold start and wrong recommendation, but also recommends the optimal product closest to the preference for the user from the perspective of the user.
In the above, the proposed multi-granularity fuzzy data fresh agricultural product recommendation algorithm under the cosine similarity method comprises the following steps:
1. using multi-granularity scoring ratio NkConstructing a multi-granularity fuzzy data processing model;
2. a multi-granularity fuzzy data processing model under a comprehensive multi-mode is synthesized to obtain a comprehensive multi-granularity effective score N with more reference valuej；
3. Calculating the similarity between the recommended fresh agricultural products and the fresh agricultural products preferred by consumers by using a cosine similarity calculation method;
4. according to the similarity
Figure BDA0002201080050000082
And sequencing the recommended fresh agricultural products, and preferentially recommending.
Secondly, calculating the similarity by using a Pearson correlation coefficient calculation method:
Figure BDA0002201080050000083
from the front to NjDefinition of (1), hereinAnd
Figure BDA0002201080050000092
can use NujAnd NvjExpression, so the formula can be changed to:
where u represents a product that the user purchased or acted upon and v represents a similar product to be recommended. N is a radical ofukAnd NvkThe ratio of the multi-granularity scores of the products u and v on a certain platform is respectively.
Figure BDA0002201080050000094
And
Figure BDA0002201080050000095
represents the average of the u, v products in different multi-particle size modes. Finally calculated correlation coefficient ρu,vThe larger the similarity between the fresh produce u and the fresh produce v representing the consumer's preference.
In the above, the multi-granularity fuzzy data fresh agricultural product recommendation algorithm under the Pearson correlation coefficient method is proposed to have the following steps:
1. using multi-granularity scoring ratio NkConstructing a multi-granularity fuzzy data processing model;
2. a multi-granularity fuzzy data processing model under a comprehensive multi-mode is synthesized to obtain a comprehensive multi-granularity effective score N with more reference valuej；
3. Calculating a correlation coefficient between the recommended fresh agricultural products and the fresh agricultural products preferred by consumers by combining a multi-granularity fuzzy data processing method and a Pearson correlation coefficient method;
4. according to the correlation coefficient rhou,vAnd sequencing the recommended fresh agricultural products, and preferentially recommending.
Detailed description of the preferred embodiment
The embodiment of the invention adopts the recent fresh agricultural product data of three big head fresh agricultural product platforms. The three big-head fresh agricultural product platforms are respectively the Beijing east fresh and the O2O fresh agricultural product electronic commerce platform of the Ali system, the Karman fresh and the Jingdong electronic commerce ecological daily superior fresh. The product data can be updated immediately by the platforms, and the timeliness and the accuracy of the acquired data are guaranteed. And (4) reasonably processing the data according to the recommendation method of the established fresh agricultural product recommendation model to finish application analysis.
In the fresh agricultural product e-commerce industry, the product data volume is huge, the user requirements are various, and the recommendation service needs to be updated continuously according to the change of the user requirements. The recommendation result is influenced by fuzzy data fed back by the user, and most platforms adopt a multi-granularity idea to process the data because the data is subjective intention of the user. For example, in three platforms of the application analysis, boxed Ma fresh is processed by a mode with the granularity of 3 for user fuzzy data evaluation, Jingdong fresh is processed by a mode with the granularity of 5, and daily Youman fresh is processed by a percentile mode, namely the mode with the granularity of 100. Therefore, the multi-granularity fuzzy data processing modes of the same product on different platforms are different, which also causes the data difference of the product. Therefore, the recommendation model established on the basis of the comprehensive multi-granularity fuzzy data processing mode by combining the real data is applied to an actual platform, and the condition of the actual application of the recommendation model is necessarily analyzed.
The application analysis acquires data of three types of fresh agricultural products preferred by users of three platforms, and the data are all derived from real data of the platforms. On one hand, the difference and the effect of the multi-granularity fuzzy data processing modes of different platforms can be obtained through comparison in data analysis; and on the other hand, the built recommendation model is subjected to application analysis by integrating the multi-platform data, the recommendation sequence of the preferred products is obtained, and the optimal products are recommended for the user.
First group, preferred products: bananas. Such products may represent most fruit products, and are diverse in origin and variety. Therefore, bananas were selected as the first preferred product for application analysis.
TABLE 1 platform data (preferred products: bananas)
Figure BDA0002201080050000101
Second group, preferred products: a fish food. The product can represent part of aquatic products and part of meat products, and has the advantages of rich varieties and high daily eating frequency. So fish are selected as a second type of preferred product for application analysis.
TABLE 2 platform data (preferred products: Fish)
Figure BDA0002201080050000102
Figure BDA0002201080050000111
Third, product preference: and (4) vegetables. Three daily-use vegetables are selected to represent most vegetable products, the vegetables are various in variety and category, and leaf vegetables, melons and fruits are selected at the moment. The application analysis is carried out on the vegetables as a third type preference product.
TABLE 3 platform data (preferred products: vegetables)
Figure BDA0002201080050000112
And collecting and summarizing three types of preference product data of the three platforms, and respectively processing the data. The data processing is mainly carried out by using a recommendation model method. The data are processed using two types of methods, namely, satisfaction and correlation coefficient. The actual recommendation effectiveness of the recommendation models constructed in the two methods is verified. And the processing result is also used for comparing the difference and the quality of the two methods, and finally, the recommendation model constructed by the method which is most suitable and has the best recommendation effect is selected for summarizing.
TABLE 4 data processing results (preferred products: bananas)
Figure BDA0002201080050000113
TABLE 5 data processing results (preferred products: Fish)
TABLE 6 data processing results (preference product: vegetables)
Figure BDA0002201080050000123
The results obtained by data processing are used for recommendation sequencing, and the recommendation sequencing generated by the recommendation models constructed by the two methods is shown in the following table. It can be seen visually that the recommendation sequences generated by the two methods in different categories are different to some extent.
TABLE 7 recommendation ranking
Figure BDA0002201080050000124
In order to analyze the recommendation effectiveness and respective advantages and disadvantages of the two methods, the recommendation model constructed by the most effective method is selected in comparison. Here, the comprehensive multi-granularity effective score N is usedjFor the basis, trend graphs are generated for the recommendation ranking in the two modes respectively, and refer to fig. 3, fig. 4 and fig. 5.
The recommendation ranking generated by the satisfaction degree method has more uncertain factors along with the trend of the ranking at the back, and many recommendations at the back are higher than those at the front, and the trend does not show a descending trend. The average comparison trend graph of the two shows that the satisfaction degree sequence accords with the normal logic, and the more backward the recommended sequence is, the more downward the recommended sequence is; but the correlation coefficient rises with the backward shift of the recommendation order, which is just contrary to normal logic. Therefore, under this application analysis, the correlation coefficients are likely to produce uncertain low quality recommendations. The satisfaction recommendation sorting modified by the cosine similarity accords with the logic of the user, and the low-quality product cannot be sorted in front.
After data of three platforms, different granularities and various products are analyzed, it can be seen that in the multi-granularity fuzzy data processing, the more the granularity quantity is, the smaller the processed user evaluation fuzzy data differentiation is, and the situation that the individual evaluation influences the whole is not easy to occur. Therefore, when multi-granularity fuzzy data are applied in the recommendation system, fuzzy data processing of user evaluation on multiple platforms and finer granularity is needed to achieve a product characteristic attribute closer to reality.
And finally, determining the first-class recommendation model as a final recommendation model for research, namely a multi-granularity fuzzy data-based fresh agricultural product recommendation model using a satisfaction method. Under the recommendation model, fresh agricultural products such as fruits, vegetables and aquatic meats are well ranked in the recommendation. Most importantly, the recommendation ranking is not only based on the similarity of the user preferred product and the recommended product, but also takes the user satisfaction into account. Therefore, the product which accords with the preference behavior of the user can be recommended in the recommendation, and the optimal recommendation based on the recommendation can be generated for the user. This will have very big realistic meaning to present fresh agricultural product electricity merchant platform, not only can improve the conversion rate, bring higher income, can also promote user experience, reinforcing user's viscosity to this is outstanding in the market that fresh agricultural product electricity merchant competition is more and more intense at present and is enclosed.
The relevant theory of the embodiment of the invention is as follows:
the idea of multi-granularity calculation can solve problems from different granularities at different levels or in different hierarchies, and thus is always one of the basic ideas for solving the actual problems. As such, multi-granularity has become one of the fastest growing information processing ideas in data applications and network-centric systems.
As a methodology, multi-granularity is mainly applied to process some uncertain, fuzzy and massive information. Based on the above, in the process of solving the problem, the granularity is used as the object to be processed for decomposition, so that the problem of fuzzy data processing is better solved. The massive fuzzy data of the processing object is put into proper granularity, so that the real and specific evaluation of the processing object can be better obtained.
In the fresh agricultural product e-commerce, the evaluation of a consumer on a product is subjective, so the product usually belongs to fuzzy data, cannot be utilized and is applied to a recommendation algorithm. At this moment, the fuzzy data needs to be quantized, and the concept of multi-granularity is adopted to enable consumers to form objective data by subjective judgment of the consumers, so that the most real consumer will is collected, and the fuzzy data of the consumers are quantized to be capable of being used in an algorithm. Today, most head-end fresh e-commerce platforms have adopted such an approach. For example, the Aries Kamameshi adopts 3 granularities to collect fuzzy data of consumer wishes, namely good comment, medium comment and bad comment, which is the most common mode of the E-commerce platform. And an independent fresh electricity commercial platform uses a good comment rate mode every day only when the fresh electricity commercial platform is good and fresh, and percentage reference is adopted. In such a mode, the granularity is more, and the data has more reference value. The Jingdong fresh food simultaneously comprises two modes, wherein in one mode, fuzzy data of consumer will are collected by 5 granularities, and in the other mode, the good evaluation rate which is the same as that of the daily excellent fresh food is adopted, namely the percentage reference is realized. The two modes are put together, two multi-granularity modes are fused, and the intention of a consumer is more accurately expressed. Therefore, the data of the product is more fit for the consumer, and the reference, recommendation, conversion and the like are more convenient.
The invention mainly researches the application of multi-granularity fuzzy data in the E-commerce recommendation algorithm of fresh agricultural products. And optimizing a recommendation algorithm on the basis, recommending high-quality fresh agricultural products which accord with the preference for the user, and improving the purchase intention and the transaction success rate of the user of the fresh agricultural product e-commerce platform.
At present, in China, the research on the recommendation algorithm is basically concentrated on research organizations in various colleges and universities. Although they have done a lot of research on recommendation systems, most are in fact in the theoretical phase, and the construction and implementation of recommendation systems is still in the phase of departure. However, in the domestic large-scale e-commerce shopping platform, such as the platform of Alibaba, Jingdong, Mei Tuo and the like, recommendation service is already carried out on the platform, and a perfect recommendation system inevitably enables the companies to win the intense market competition. Although the application of the recommendation algorithm is a lot of research, the common one can be briefly summarized as the following:
(1) content-based recommendation algorithm
The user's behavior information is a data source for a content-based recommendation algorithm. Extracting characteristics of the product which is subjected to the interaction behavior by the user, and establishing a preference characteristic model for the user based on the characteristics of the user interaction product, so as to recommend other products similar to the characteristics of the product which is subjected to the preference by the user. In practical applications, content-based recommendation algorithms require an appropriate model to construct product features and compare similarities between a user's preferred product and candidate products. The method mainly comprises the following three steps:
1. extracting characteristic attributes of each product to represent the product;
2. constructing a preference characteristic model of the user by utilizing the characteristic attribute data of the product which the user likes (and dislikes);
3. and comparing the user preference characteristics obtained in the second step with the characteristics of the products to be recommended, and calculating the similarity of the characteristics of the user preference products and the candidate products, wherein high similarity represents that the user probably likes the products, and low similarity represents that the user does not like the products.
The specific calculation method of cosine similarity comprises the following steps:
where c represents user preferences, s represents the products to be recommended to user c, and u (c, s) represents the degree of recommendation of product s to user c.
In the content-based recommendation algorithm, the preference characteristics of each user are extracted according to the behavior information of the user, and different users are independent of each other. Moreover, the accuracy of recommendation can be continuously improved by continuously increasing the dimensionality of the feature extraction of the user and the product;
however, there are disadvantages in that the feature extraction of the product is relatively difficult, and in many cases, it is difficult to effectively extract more data, and the extracted feature attributes of the product can only represent some aspects of the product and not all aspects of the product in almost all practical cases. Similarly, the algorithm recommendation result is based on the past behavior information of the user, so that only products similar to the past preference products of the user are recommended, and new interesting products cannot be discovered for the user. When a new user enters the platform, the preference characteristics of the new user cannot be acquired due to no historical behavior, and in this case, the algorithm is invalid.
(2) Hybrid recommendation algorithm
The idea of popular adoption is needed regardless of daily learning or algorithm use, and the mixed recommendation algorithm is the popular adoption algorithm. The hybrid recommendation algorithm combines the advantages and disadvantages of the multiple recommendation algorithms through the combined use of the multiple recommendation algorithms, and makes the best of the advantages and disadvantages or compares the advantages and disadvantages with each other, so that the best recommendation effect is achieved. To this end, researchers have developed various mixing strategies. According to Burk's classification criteria, the hybrid recommendation algorithm has seven hybrid strategies, which can be summarized in three basic methods, respectively: bulk mixing, parallel mixing and linear mixing. The hybrid recommendation algorithm can be combined with the advantages of various recommendation algorithms, and the defects of various algorithms are effectively avoided. But also combining multiple recommendation algorithms would be more complex.
(3) Collaborative filtering recommendation algorithm
1. Collaborative filtering algorithm based on users
Generally in user-based collaborative filtering algorithms, a user is considered to be interested in products purchased by other users that have similar preferences as his interests. The collaborative filtering algorithm based on the user is mainly divided into two modules, wherein the user preference characteristics are extracted, the user preference similarity is calculated, and the preference degree of the user for recommended products is calculated based on the user similarity. Specifically, for users u and v, where n (u) represents the product that user u purchased, and n (v) represents the product that user v purchased. Therefore, the interest preference similarity of user u and user v can be defined by the following formula:
Figure BDA0002201080050000161
wherein | n (u) andu (v) | represents a common product collection purchased by users u and v, and | n (u) @ n (v) | represents a union of product collections purchased by users u and v. Then, the following formula is used to define the preference degree of the user u for the product l:
wherein s (u, m) represents m users with the highest similarity to user u, wuvRepresenting the similarity of interest preferences between users u and v, rvlRepresenting the interest preference of the user v for the product l, and N (l) represents the set of users who have bought the product l.
2. Collaborative filtering algorithm based on products
In product-based collaborative filtering algorithms, it is generally assumed that a user will be interested in similar products to the products he has purchased. The collaborative filtering algorithm based on the products is also mainly divided into two modules, wherein firstly, similarity calculation is carried out on the attribute characteristics among the products, and secondly, a recommended product list is generated by combining the interest preference of the users on the products and the similarity of the attribute characteristics among the products.
The method for calculating the similarity of the products comprises the following steps:
cosine similarity:
Figure BDA0002201080050000171
euclidean distance:
Figure BDA0002201080050000172
Figure BDA0002201080050000173
cosine similarity:
Figure BDA0002201080050000174
pearson correlation coefficient:
tanimoto coefficient:
Figure BDA0002201080050000181
the collaborative filtering recommendation algorithm has the main advantages that the content does not need to be deeply mined, only the behavior preference of the user is relied on, and the unstructured complex object can be processed. The algorithm principle is simple and the application range is wide. The product-based collaborative filtering algorithm may have reasons for recommendations after generating a recommendation list. Has good explanatory property. The collaborative filtering algorithm based on the user can deeply analyze the user requirements, and after the recommendation result is generated, the preference characteristics of the user are relatively more met. However, user-based collaborative filtering algorithms rely heavily on user preference profile data. At the same time, cold start and product overheating are also problems that are often associated with collaborative filtering algorithms.
In summary, through research and analysis of several commonly used recommendation algorithms, it is found that the principle of the collaborative filtering recommendation algorithm is more suitable for the objective of the present invention, which is to recommend a desired product for a user, and to recommend a suitable fresh agricultural product for the user. Through the analysis, the recommendation model constructed based on the multi-granularity fuzzy data applied to the collaborative filtering recommendation algorithm is more suitable.
The above disclosure is only for a few specific embodiments of the present invention, however, the present invention is not limited to the above embodiments, and any variations that can be made by those skilled in the art are intended to fall within the scope of the present invention.

Claims (5)

1. A fresh agricultural product recommendation method based on multi-granularity fuzzy data is characterized by comprising the following steps:
acquiring characteristic parameters of a user and a product;
constructing a characteristic model according to the characteristic parameters, and determining user preference parameters according to the characteristic model;
and inputting the user preference parameters into a multi-granularity fuzzy data recommendation model to obtain a recommendation result, and feeding back the recommendation result to the user.
2. The method of claim 1, wherein the method of recommending fresh produce based on multi-granularity fuzzy data,
the user characteristic parameters comprise: age, academic history, income, and marital status;
the product characteristic parameters comprise: product quality, price and logistics time.
3. The method for recommending fresh agricultural products based on multi-granularity fuzzy data as claimed in claim 1, wherein the inputting of the user preference parameter into the multi-granularity fuzzy data recommendation model and the determining of the recommendation result specifically comprise:
constructing a multi-granularity fuzzy data processing model by adopting multi-granularity grading proportion;
synthesizing a multi-granularity fuzzy data processing model under multiple modes to obtain a comprehensive multi-granularity effective score;
calculating the similarity between the recommended product and the user preference product according to the satisfaction degree of the user to the recommended product;
and sequencing the recommended products according to the similarity to generate a recommendation result.
4. The multi-granularity fuzzy data-based fresh agricultural product recommendation method according to claim 3, wherein the multi-granularity score is proportional to the expression:
Figure FDA0002201080040000011
the expression of the comprehensive multi-granularity effective score is as follows:
Figure FDA0002201080040000021
wherein, XatScore for t total data at granularity b, NkmThe multi-granularity score ratio under m different multi-granularity modes is shown, t is the total number of data pieces, and m is the number of different multi-granularity modes.
5. The method as claimed in claim 4, wherein the satisfaction degree P of the user on the recommended product is determined by the multi-granularity fuzzy datavThe expression is as follows:
wherein N ismaxA comprehensive multi-granularity effective score for the user's preferred ideal product; n is a radical ofkvAnd v is the ratio of the multiple particle size scores of the v product, and v is the v product.
CN201910865268.4A 2019-09-12 2019-09-12 Fresh agricultural product recommendation method based on multi-granularity fuzzy data Active CN110634051B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201910865268.4A CN110634051B (en) 2019-09-12 2019-09-12 Fresh agricultural product recommendation method based on multi-granularity fuzzy data
ZA2020/05623A ZA202005623B (en) 2019-09-12 2020-09-10 A fresh agricultural product recommendation method based on multi-granularity fuzzy data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910865268.4A CN110634051B (en) 2019-09-12 2019-09-12 Fresh agricultural product recommendation method based on multi-granularity fuzzy data

Publications (2)

Publication Number Publication Date
CN110634051A true CN110634051A (en) 2019-12-31
CN110634051B CN110634051B (en) 2022-04-12

Family

ID=68971139

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910865268.4A Active CN110634051B (en) 2019-09-12 2019-09-12 Fresh agricultural product recommendation method based on multi-granularity fuzzy data

Country Status (2)

Country Link
CN (1) CN110634051B (en)
ZA (1) ZA202005623B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111859635A (en) * 2020-07-03 2020-10-30 中国人民解放军海军航空大学航空作战勤务学院 Simulation system based on multi-granularity modeling technology and construction method
CN111899063A (en) * 2020-06-17 2020-11-06 东南大学 Fresh agricultural product online recommendation method considering customer consumption behaviors and preference

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100082359A1 (en) * 2008-09-30 2010-04-01 Yahoo! Inc. Multi-Granular Age Range Products For Use in Online Marketing
CN109711728A (en) * 2018-12-27 2019-05-03 陕西师范大学 The double-deck multiple target power dispatching method based on electric power uncertainty and low-carbon demand
CN110135587A (en) * 2019-05-10 2019-08-16 山西大学 It is hesitated based on section and obscures the Multiple Attribute Group Decision of more granularity decision rough sets

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100082359A1 (en) * 2008-09-30 2010-04-01 Yahoo! Inc. Multi-Granular Age Range Products For Use in Online Marketing
CN109711728A (en) * 2018-12-27 2019-05-03 陕西师范大学 The double-deck multiple target power dispatching method based on electric power uncertainty and low-carbon demand
CN110135587A (en) * 2019-05-10 2019-08-16 山西大学 It is hesitated based on section and obscures the Multiple Attribute Group Decision of more granularity decision rough sets

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
J.A. MORENTE-MOLINERA 等: "On multi-granular fuzzy linguistic modeling in group decision making problems: A systematic review and future trends", 《KNOWLEDGE-BASED SYSTEMS》 *
陈秀明: "基于多粒度犹豫模糊语言信息的群推荐方法研究", 《中国优秀博硕士学位论文全文数据库(博士)信息科技辑(月刊)》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111899063A (en) * 2020-06-17 2020-11-06 东南大学 Fresh agricultural product online recommendation method considering customer consumption behaviors and preference
CN111859635A (en) * 2020-07-03 2020-10-30 中国人民解放军海军航空大学航空作战勤务学院 Simulation system based on multi-granularity modeling technology and construction method

Also Published As

Publication number Publication date
CN110634051B (en) 2022-04-12
ZA202005623B (en) 2021-06-30

Similar Documents

Publication Publication Date Title
CN106651546B (en) Electronic commerce information recommendation method oriented to smart community
CN102542474B (en) Result ranking method and device
CN111932336A (en) Commodity list recommendation method based on long-term and short-term interest preference
CN102609523A (en) Collaborative filtering recommendation algorithm based on article sorting and user sorting
CN102411754A (en) Personalized recommendation method based on commodity property entropy
CN101482884A (en) Cooperation recommending system based on user predilection grade distribution
CN103559622A (en) Characteristic-based collaborative filtering recommendation method
CN108334592A (en) A kind of personalized recommendation method being combined with collaborative filtering based on content
CN110634051B (en) Fresh agricultural product recommendation method based on multi-granularity fuzzy data
CN104298787A (en) Individual recommendation method and device based on fusion strategy
CN109508407A (en) The tv product recommended method of time of fusion and Interest Similarity
Levinas An analysis of memory based collaborative filtering recommender systems with improvement proposals
CN105809275A (en) Item scoring prediction method and apparatus
Tait et al. Valuing preferences for environmental sustainability in fruit production by United Kingdom and Japanese consumers
CN115809926A (en) Innovative alliance distribution cooperation transaction platform for converging alliance resources of multiple distribution channels
CN116402577A (en) Transaction information reading method and system based on online shopping
CN116503142B (en) Partner intelligent marketing scheduling data processing system
US20220383344A1 (en) Generating numerical data estimates from determined correlations between text and numerical data
CN112381627B (en) Commodity scoring processing recommendation method and device under child-care knowledge
CN110348964A (en) It is a kind of based on the wisdom electronic commerce recommending method more perceived
CN116187808A (en) Electric power package recommendation method based on virtual power plant user-package label portrait
CN111435514A (en) Feature calculation method and device, sorting method and device, and storage medium
CN114997959A (en) Electronic intelligent product marketing recommendation method
CN110163696A (en) A kind of electric business shopping air navigation aid and its system based on big data
CN111061943A (en) Dish recommendation system and dish recommendation method based on data mining and analysis

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant