KR20200097884A - A method for recommending an item based on behavior analysis of a group with similar preferences - Google Patents

Abstract

The present invention relates to a system capable of recommending an item when a new user shows similarity in groups, preferences and the like by forming similar collectivization in accordance with the preferences of the user in a social network environment including the internet, and a method thereof. The present invention provides a scientifically reliable, fast and efficient system. To this end, the present invention considers anthropological factors and marketing factors when selecting the recommended item, uses an artificial neural network, and adopts polar coordinates for flexible grouping.

Description

A method for recommending an item based on behavior analysis of a group with similar preferences

The present invention relates to a system and method for recommending at least one item to a consumer on the basis of a cluster showing a similar tendency according to the preferences of consumers.

In the past 2 to 30 years ago, in a supplier-oriented society, the supply of everything was short, so if a supplier supplied a specific product, the consumer had no choice but to buy regardless of the price high or low. However, as times have changed, the Internet and social networks are active, and consumers' diverse desires and demands are diversified, the market has thoroughly grasped consumer needs and completely refined the marketing method accordingly.

In the era of mass production of a large number of items, such as in recent years, understanding the consumption tendencies or preferences of specific individuals and recommending items suitable for that specific person provides various temporal economic benefits to consumers or suppliers.

In the present invention, an automated collaborative filtering (ACF) system has been proposed in which each item released on the market in a specific field is classified according to the user's preference and recommended to users showing similarity.

ACF is a term proposed in the sense that consumers cooperate in filtering specific items by recording their preferences and is also referred to as a recommender system. However, the former and the latter differ in the method of calculating the similarity value, and there is a difference between using the rating (ratin g) or the probability (proablity) when calculating the distance according to the similarity, but both calculate the similarity by distance. It is known that there is a problem that a lot of load is applied when calculating (high computation cost).

Accordingly, according to the Jester 2.0 paper (Jester 2.0) proposed by Ken Goldberg et al. of Berkeley University published at the ACM SIGIR99 Society (August 19, 1999), a principal component analysis (PCA) is a more effective method. , Hereinafter referred to as PCA) and clustering-based algorithms are proposed. Therefore, since this Gester paper does not classify preferences by distance, it is based on a group, so it is a more advanced proposal.

However, in these Jester thesis, the probability of the recommendation system is increased by the introduction of a group by the PCA method, but this is proposed for the humor that can collect as many as tens of thousands of data on the web. Likewise, it is not suitable for targeting not so much data. In addition, according to the Jester thesis, there is a problem in that a specific consumer cannot recommend an item because the preference of a specific consumer cannot be grasped in a state in which at least 5 or 6 items are not selected and graded.

In addition, X-Y coordinates are used to divide or subdivide groups into two dimensions, but this has a problem in that there may be a certain deviation from the distance from the concentration center, that is, close to preference.

The present invention has been devised to overcome the problems of the above papers, and an object of the present invention is to provide a more statistically reliable item recommendation system. Another object of the present invention is to provide an item recommendation system capable of reducing processing capacity and processing speed. Another object of the present invention is an item that can recommend an item even if a specific consumer's preference for one item is identified. It is to provide a recommendation system.

The recommendation system and its recommendation method according to the present invention are related to the distance from the center center by converting the grouping by XY coordinates into polar coordinates, so that the difference between groups can be more meaningfully set, and it is advantageous to reset groups, etc. , In the XY coordinates, a number of X and Y coordinates must be memorized, but since the polar coordinates are easily set as r and θ, there is a side effect of reducing the memory capacity and operation time of the program. In addition, by considering actual demographic factors and marketing factors, there is an advantage of increasing the reliability of the recommendation set with less data. In addition, by applying the artificial neural network, it is possible to recommend an appropriate item with a minimum evaluation score, which has the advantage of increasing the accuracy of the artificial neural network further as the data increases.

1 is a flowchart showing a process actually applied on the web by applying the method according to the present invention

Briefly explaining the recommendation system according to the present invention, basically, this recommendation system is based on communication, especially under the WWW environment of the Internet, and the consumer is given two sets. One is a predictor set and the other is a recommendation set. Here, the prediction set may also be called an evaluation set. When a new consumer ranks each item in the prediction set, the recommendation system lists the recommended item suitable for it in the recommendation set, and this recommendation item is expected to receive a high rating from the new consumer.

First, a method of setting a prediction set will be described. The prediction set selects product items that can have various variance distributions and collects ratings, that is, the number of evaluation points, from multiple consumers. This level is, for example, 5 levels, and can be divided into -2, -1, 0, 1, 2, etc. In other words, if only N items that are expected to have a large variance in preference are selected and preference is received by M evaluators, the matrix is composed of M × N.

Analyzing this matrix excludes rows or columns with low variance, for example, items all rated near 2, all rated near -2, and evaluators who gave all items a similar rating. This is because they are items and evaluators that are difficult to judge a particular person's preferences. At this time, the number of selected items is advantageous in that it is advantageous in that it can reduce the evaluation time and make the evaluator not bored, but it is of course possible to use a method of showing them on the screen one by one. . Next, a correlation matrix between items in the selected M'× N'matrix is obtained from the equation below. The correlation matrix is a matrix of the correlation coefficient between a specific item and other items.

Here, Xi and Yi refer to the rating of the i-th evaluator, and X(bar) and Y(bar) refer to the average of each X and Y values. This coefficient is a number between -1 and 1. If it is close to 1, it is interpreted as having a large correlation, if it is close to -1, it is interpreted as having the opposite correlation, and if it is 0, the correlation is Is interpreted as missing

Next, the artificial neural network is operated, and the artificial neural network is divided into a learning stage and a production stage. That is, in the learning stage, the evaluation score of each recommender is input to the input pattern, and a cluster of each recommender is set to the output pattern. In this process, the weight of the artificial neural network is adjusted according to the learning algorithm according to the error between the output pattern calculated in a series of input/output patterns and the set output pattern, and when the learning continues and the error falls within the allowable range. When the learning is finished and the learning is finished, the weight is memorized. In the production stage, when an input pattern to be actually applied is supplied, an output pattern is derived by the learned artificial neural network. In other words, in the learning stage, input and output form a pair. The more learning patterns, the more advanced the artificial neural network can learn, and the artificial neural network once learned can predict the output only with the input pattern. Therefore, the artificial neural network in the state of completion of learning predicts the output pattern with only one calculation without the need for iteration to correct errors. Since the concept of such an artificial neural network is conventional, a more detailed description will be omitted.

As described above, after learning of the recommendation system, grouping, and artificial neural network is completed, those with an evaluation score greater than a certain value from evaluators belonging to each cluster are extracted as a recommendation list. The recommendation list can be selected by the average evaluation value or the item with a high evaluation score can be selected. The first recommendation system according to an embodiment of the present invention is completed by the method and system described above, and when a new consumer evaluates some of the items listed in the prediction set, it is divided into polar coordinates on the eigenspace according to the artificial neural network. A specific group is defined and a list of recommendations from that group is provided. At this time, the consumer can continue to evaluate the rest of the evaluation items, and the person who has evaluated all of the predicted item sets becomes the recommendation source, and the user's evaluation and purchase affect the recommendation set.

Next, in this new recommendation set, evaluators are newly constructed, and demographic and marketing factors are considered for precise recommendation. Demographically, the recommendation list is extracted by reorganizing it by age, gender, and income, which affects actual purchase, or recently evaluated in consideration of time recency, frequency, and actual purchase. An additional score is added to the evaluation scores of items and items with high purchase frequency, and the group is newly grouped and a new recommendation set is created. This is adopted in the present invention in order to perform more precise recommendations with less data because the field to which the present invention is applied is less data than the nanny applied in the Jester paper.

In other words, when describing cosmetics, for example, colors or tastes that you like may differ according to age, and the price and actual purchase or not should be the main judgment factors. In other words, the error of recommending cosmetics that teenagers like in their 40s can be corrected. This method can be achieved with a program capable of automatically upgrading the recommendation set at a predetermined period.

In the case of the Jester paper, XY coordinates should be fully readjusted. However, since polar coordinates are used in the present invention, group subdivision or subdivision is performed by adjusting θ and r to subdivide groups from 40 groups to 80 groups. Can be done easily.

On the other hand, as for the prediction set of the primary recommendation system, a number of new recommendation sources are registered, evaluation data is accumulated, and new items are increased. It is determined after a predetermined time elapses after driving the drive. This time may be judged as the number of items evaluated. Accordingly, a second recommendation system is completed in which a new recommendation set, a new grouping, upgrade of the recommendation list, and learning of the artificial new network are performed.

This secondary recommendation system repeats the steps of setting up a new recommendation set as described above after a predetermined period of time has elapsed, and when more than a certain amount of data is accumulated, a third recommendation system is formed again. . 1 is a flowchart illustrating a process of receiving a recommendation by a consumer on a web in which a method according to the present invention is set as described above.

In other words, if the specific consumer first judges whether there are items that have not been evaluated in the predicted set, and if all items are evaluated, the recommended items are listed in the recommended set. If not, the predicted set is asked whether to evaluate the item and attempts to evaluate it. List the items in the list and evaluate them, and if they say not to evaluate them, they list recommended items.

When an item is evaluated in the prediction set, the artificial neural network is driven to determine the group to which the specific consumer belongs, and recommended items are listed in the recommendation set.

Of course, as shown in the flowchart, that particular consumer can continuously evaluate the items in the prediction set.

Claims (1)

-Evaluating a plurality of pre-selected evaluators for each of a plurality of preset items;
-Grouping the plurality of evaluators into individual groups according to the correlation between the evaluated plurality of items;
-Setting an item that received high evaluation from the group classified as the individual group as a recommended item of the individual group;
-Setting a relationship between the evaluation of the item and the set group by using the evaluation of each item and the set group;
-Evaluating at least some of the plurality of items by a specific consumer;
-Determining a specific group to which the specific consumer belongs using the relationship determined in the step of establishing a relationship between the item and the group based on the evaluation of the item evaluated by the specific consumer;
-Item recommendation method comprising the step of recommending the recommendation item of the specific group to the specific consumer

Images