JP2007041971A - Recommendation system, recommendation method and recommendation program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a recommendation system that can recommend more suitable merchandise even to consumers with no purchase history. <P>SOLUTION: The system, which recommends one or more articles of merchandise out of a merchandise group comprising a plurality of competitive articles, has a merchandise database storing, for each article, attributes, and levels indexing the degree of influence on purchasing power, means for inputting an importance level to each attribute, means for translating the input importance level into a factor, means for multiplying the indexing level by the factor-translated importance level to produce a weighted indexing level, means for calculating a distance between a virtual article and each article from differences between the weighted indexing levels of a consumer-desired article and each article, and means for displaying one or more articles as recommended articles in ascending order of calculated distance. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

The present invention relates to a recommendation system, and more particularly, to a system for recommending an optimal product to a consumer according to a consumer's request. In this specification, “product” is used in a broad sense, and “product” includes not only tangible goods but also intangible services and information. The recommendation system according to the present invention is executed on a computer network or by using an offline terminal installed in a store.

With the spread of the Internet, the number of consumers who purchase products through the Internet is increasing, and so-called IT-based product recommendation is also actively performed. Conventionally, several methods are known as recommendations.

Questionnaire-based recommendation is to request gender, age, occupation, hobby interest, etc. when registering as a member and recommend products according to the declaration. Questionnaire-based recommendations are the most basic recommendations, but have drawbacks such as limited information input by consumers, less information maintenance, and inability to accommodate new categories.

Rule-based recommendation is a method for recommending products based on rules that have already been verified based on purchase history and access history. For example, a rule based on IF THEN is set, such as recommending snacks to people who bought beer.

Real-time recommendations use the purchase history and access history in the same way as the rule base. However, real-time recommendations analyze them statistically, and the system automatically selects products suitable for consumers on the WEB. Recommend in real time. There are the following two types depending on the algorithm. One is collaborative filtering, which is a recommendation algorithm based on the idea that people with the same preference will purchase the same thing. The other is content-based filtering, which is a method of constructing a deep knowledge structure model for content such as products, and observing and learning individual user responses to the model to realize optimal recommendation.

The following applications are exemplified as prior documents related to recommendation or recommendation.
JP 2002-279279 A JP 2002-334257 A JP 2003-150835 A JP 2004-21810 A JP 2004-318597 A

Conventional recommendations mainly use purchase histories and access histories, and cannot make effective recommendations for consumers who do not have purchase histories or access histories. In many cases, recommendations are used mainly by consumers who select products and recommend them to consumers in order to encourage consumers to purchase, and are used passively by consumers. It is done. On the other hand, there are many cases where a consumer needs to recommend a product even when he / she wants to actively purchase the product. In particular, when a wide variety of competing products (even if they have the same attributes, each with different levels) are flooded, and consumers want to get a certain product, how to select what they want efficiently Is becoming important. That is, it is important how to link what the consumer desires with the offered product, and a system is also required for the consumer to recommend a product suitable for his / her wishes, desires or conditions.

An object of this invention is to provide the recommendation system which can recommend more optimal goods also to the consumer with no purchase history. Another object of the present invention is to provide a product recommendation system that can accurately recommend a product that meets the consumer's wishes from a plurality of products that are actually provided.

The technical means adopted by the present invention is a system for recommending one or a plurality of products from a group of products consisting of a plurality of competing products, and indicates the degree of influence on the attributes and purchase of each product. The product database storing the indexed level, means for inputting the importance for each attribute, means for coefficientizing the input importance, and the coefficient indexed to the indexed level Assuming that the consumer wants a virtual product composed of the means to obtain the weighted indexing level by multiplying the degree and the highest value of the weighted indexing level of each attribute, the virtual product and each product in each attribute Means for calculating the distance between the virtual product and each product using the difference between the weighted indexing levels, and means for displaying one or a plurality of products as recommended products in ascending order of the calculated distance. Having a recommendation system That. In addition, another technical means adopted by the present invention affects a computer to recommend one or more products from a group of products consisting of a plurality of competing products, and affects the attributes and purchase of each product. A product database storing levels indexed to indicate the degree, means for inputting the importance for each attribute, means for coefficientizing the input importance, and the coefficient for the indexed level Assuming that a consumer wants a virtual product composed of a means for obtaining a weighted indexing level by multiplying the weighted importance and a maximum value of the weighted indexing level of each attribute, the virtual in each attribute Using the difference between the product and the weighted indexing level of each product, a means for calculating the distance between the virtual product and each product, and displaying one or more products as recommended products in ascending order of the calculated distance As a means to A computer program for causing. Furthermore, another technical means adopted by the present invention is a method of recommending one or a plurality of products from a group of products consisting of a plurality of competing products, and the level of the attribute of each product has an influence on the purchase. A step of indexing to indicate the degree of giving, a step of inputting the importance for each attribute, a step of coefficientizing the input importance, and multiplying the indexed level by the coefficient to weight Assuming that the consumer wants a virtual product consisting of the step of obtaining the indexing level and the highest value in the weighted indexing level of each attribute, the virtual product and the weighted indexing level of each product are used to A step of calculating a distance between the product and each product, and a step of displaying one or a plurality of products as recommended products in ascending order of the calculated distance. In a preferred embodiment, “indexation” and “coefficientization” convert “level” and “importance” into a range of 0 to 1, respectively.

In a preferred aspect, the display means further displays a deviating attribute together with the recommended product. In one aspect, a threshold is set for the distance between the weighted indexing levels, and when the distance between the weighted indexing levels is larger than the threshold, the attribute is set as a deviating attribute. In that case, all the attributes that are different may be displayed, or only a predetermined number of attributes may be displayed in order of increasing distance. Alternatively, in another aspect, a predetermined number of attributes may be displayed in descending order of the distance between the weighted indexing levels.

The recommendation system has means for indexing the level of each product to indicate the degree of influence on the purchase. The indexing means has means for normalizing the level. Further, the indexing means has means for optimizing the normalized level. Examples of the optimization means include normalized level nonlinear conversion and normalized level dynamic range optimization. The indexing method can be changed depending on the attribute, or may be changed by the consumer. The level indexing is not necessarily a linear transformation, and may be an exponential or quadratic nonlinear transformation.

The input means includes a slider bar displayed on the display unit, a knob movably provided on the slider bar, and means for selecting a position of the knob on the slider bar, and the position of the knob on the slider bar. Corresponding to the importance coefficient is determined between 0 and 1. In one embodiment, the slider bar is divided into a plurality of sections, and a coefficient is assigned to each section, and the position of the knob on the slider bar is converted into a coefficient assigned to the section to which the slider bar belongs. Further, in one aspect, coefficients are set such that one end section corresponds to 0 and the other end section corresponds to 1, and the section between them is converted into a coefficient nonlinearly. In another aspect, coefficients are calculated so that one section corresponds to 0, and the other sections are converted into coefficients nonlinearly so that the coefficients increase toward the other end.

In one aspect, the recommendation system includes means for inputting a narrowing condition by using a part of the attribute or / and a level of the partial attribute as a narrowing condition for the product. In this case, in one aspect, the virtual product is configured from the highest value in the weighted indexing level of each attribute of the product after narrowing down. Further, even when narrowing down, the virtual product may be configured from the highest value in the weighted indexing level of each attribute of the product before narrowing down.

In the present invention, since the level of each attribute is indexed and the importance of each attribute is coefficientized for each consumer, it is possible to accurately select a product that meets the consumer's wishes from a plurality of actually provided products. Can recommend. Since attribute importance level information is acquired for each consumer, a more optimal product can be recommended even for a consumer who has no purchase history.

Further, by displaying the divergence attribute together with the recommended product, it is possible to select a product to be purchased using the divergence status as a clue from a plurality of recommended products. For example, even if a product with a small overall distance from a virtual product (ideal product) has an unacceptable divergence in a specific attribute, the overall distance is slightly larger. It is also possible to select a product having a small distance in the attribute.

[A] Overall Configuration of the System of the Present Invention A recommendation system according to the present invention includes a product database storing attributes and indexed levels of each product, means for inputting importance for each attribute, and input importance Means for coefficientizing the degree, means for multiplying the indexed level by the coefficientized importance to obtain a weighted indexing level, and the highest value in the weighted indexing level of each attribute. Assuming that the virtual product is a product that the consumer wants, a means for calculating the distance between the virtual product and each product using the weighted indexing level of the virtual product and each product, and one in ascending order of the calculated distance Or a means for displaying a plurality of products. The recommendation system according to the present invention includes one or a plurality of computers, and the computer includes processing means, storage means, input means, output means or display means, a control program for operating the computer, and the like. .

In one aspect, the system of the present invention includes an EC server and a client terminal as shown in FIG. 1, and both are connected via the Internet exemplified as a computer network. Both the client terminal and the server have a basic configuration as a computer (processing unit, storage unit, input unit, output unit or display unit, control program for operating the computer, and the like). In addition, the client terminal and the server are provided with transmission / reception means that enables mutual exchange of information via a computer network.

As will be described later, various product databases are stored in the storage means of the server. The product database is processed and calculated by normalization means, optimization means stored in the storage means of the server, narrowing-down means input from the input means of the client terminal, and means for coefficientizing the importance of the attribute. It is displayed on the display means of the client terminal. In another aspect, the system of the present invention may be configured by an offline terminal installed at a store instead of on a computer network.

[B] The product database is stored in the storage means of the product database server. A product is defined by a plurality of product attributes and levels. For example, as shown in FIG. 8, regarding the product A which is a digital camera, the product attributes include price, number of pixels, optical zoom, size, weight, manufacturer, etc., and the price is 28000 yen, The number of pixels is 3 million, the optical zoom is 3 times, the size is 20 × 50 × 40, the weight is 230 g, and the manufacturer is ABC. Not all levels need to be quantified. For example, in the attribute of manufacturer, the name of each manufacturer may be used as the standard. Alternatively, the attribute “manufacturer” may be grouped based on some standard (for example, scale), and the group number may be used as a level.

[C] In the processed product database of the product database, the level of each attribute is indexed to indicate the degree of influence on the purchase. The level is indexed by normalizing the level and optimizing the normalized level appropriately.

例えば、画素数が３０万〜８００万まであったとすると、１００万→０、８００万→１となるように計算する。また、価格のように、数値が上昇することが、買いたさにネガティブな影響を与える場合には、最高額の値を０、最低額の値を１となるように正規化してもよい（尚、この処理を最適化として行ってもよい）。 [C-1] Normalization of level normalization data is to normalize the level to a value of 0 to 1 using equation (1) as shown in the upper diagram of FIG.

For example, if the number of pixels is 300,000 to 8 million, calculation is performed so that 1 million → 0 and 8 million → 1. In addition, when the increase in the numerical value, such as the price, has a negative influence on the purchase, normalization may be performed such that the maximum value is 0 and the minimum value is 1. Note that this processing may be performed as optimization).

を用いて、正規化された水準を変換する。 [C-2] Optimization of the optimization level is exemplified by non-linear conversion of data. As shown in the lower diagram of FIG. 3, when the degree of purchase F (X) is not proportional to X, the level is nonlinearly converted. For example, when the degree of purchase is monotonously increasing, in one aspect, the expression (2),

Is used to convert the normalized level.

を用いて、正規化された水準を変換する。 When the degree of purchase is monotonously decreasing, in one aspect, Equation (3),

Is used to convert the normalized level.

例えば、価格に関して、その範囲を０〜２０万円にしたい場合、０円→１、２０万円→０となる。すなわち、１万円→０．９５、１０万円→０．５となるように調整する。そうすることで、当該データベースの最小、最大を超えた数字が追加された時の連続性を保つことができる。 Another example of level optimization is optimization of the dynamic range of data. As shown in the upper diagram of FIG. 4, when there is a problem in setting the maximum value in a certain database to 1 and the minimum value to 0, use formula (4) to provide blank space at the top and bottom. Optimize dynamic range.

For example, when the price is desired to be 0 to 200,000 yen, 0 yen → 1, 200,000 yen → 0. That is, adjustment is performed so that 10,000 yen → 0.95, 100,000 yen → 0.5. By doing so, it is possible to maintain continuity when numbers exceeding the minimum and maximum of the database are added.

を用いることもできる。 Another example of optimization is the non-linear transformation of the data (if there are peaks)

Can also be used.

Data optimization will be described in more detail. FIG. 5 illustrates a case where “the purchase level increases as the level value increases”. Examples of attributes having such a level include the number of pixels, resolution, and battery life. The left figure of FIG. 5 shows the correspondence between the normalization coefficient and the optimization coefficient. This will be specifically described with reference to the right diagram of FIG. 5, taking the number of pixels as an example. First, the number of pixels is normalized to an index between 0 and 1. Next, the normalized exponent is nonlinearly transformed using equation (2) (a = 0.5). Then, the non-linearly transformed exponent is optimized. Here, the dynamic range is optimized to 0.10 to 0.80 using Equation (4).

FIG. 6 illustrates a case where “the purchase increases when the value of the level decreases”. Examples of attributes having such a level include price, weight, and thinness. The left diagram in FIG. 6 shows the correspondence between the normalization coefficient and the optimization coefficient. This will be specifically described with reference to the right diagram of FIG. 6 taking the number of pixels as an example. First, the number of pixels is normalized to an index between 0 and 1. Next, the normalized exponent is nonlinearly transformed using equation (3) (a = 2). Then, the non-linearly transformed exponent is optimized. Here, the dynamic range is optimized to 0.10 to 0.90 using Equation (4).

FIG. 7 illustrates a case where the level value is the optimum value and the purchase level has a peak. As an attribute having such a level, size is exemplified. The left diagram in FIG. 7 shows the correspondence between the normalization coefficient and the optimization coefficient. This will be specifically described with reference to the right diagram of FIG. 7 taking the weight (here, it is assumed that there is an optimum value between the minimum value and the maximum value of the level) as an example. First, the number of pixels is normalized to an index between 0 and 1. Next, the normalized exponent is optimized (non-linear transformation) using equation (5) (b = 0.31).

[D] Refinement In a preferred form, a part of the attribute is used as the refinement condition. The attribute level used as the narrowing-down condition is a numerical level (for example, an upper limit price) or / and a non-numerical level (for example, a manufacturer name). The information on the narrowing condition is input from the client terminal, and the input information is transmitted to the server via the network, and the server having received the information is narrowed down to products having a level that matches the narrowing condition.

[E] A weighted product of attributes is considered to be defined by a plurality of attributes and levels. The value of each attribute depends on the consumer's weight. Regardless of whether or not there is weighting, products that are all satisfied are desired to be purchased, and when there are items that are not partially satisfied, the smaller the weighting, the smaller the effect. The ones that have the least divergence from the ones that are all satisfied are the ones consumers want to buy. Even if the deviation from the ideal point is the same 30%, if the weight is half, the contribution rate is half. Therefore, in order to recommend a product desired by the consumer, it is necessary to measure the importance of the attribute defining the product.

The importance of the attribute is determined by input from the consumer. Specifically, a screen for inputting the importance level of the attribute is displayed on the display unit of the client terminal, and the importance level is input from the input unit. The input importance is converted into a numerical value between 0 and 1 by a predetermined conversion operation. In one aspect, the importance of an attribute is input using a slider displayed on the display screen. The slider is composed of a slider bar and a knob that can be moved freely on the slider bar. The position of the knob on the slider bar represents the importance of the attribute, and each slider is moved on the slider bar. Enter the importance of the attribute. The importance of the input attribute is converted into a coefficient according to the position of the knob on the slider bar. In one form, the slider bar is divided into a plurality of sections, and a coefficient is assigned to each section, and the position of the knob on the slider bar is converted into a coefficient assigned to the section to which the slider bar belongs.

In the present invention, a non-linear slider is used as one preferred embodiment. The non-linear slider will be described with reference to FIGS. Both the slider bars in FIGS. 15 and 16 are divided into seven sections, and a coefficient is determined corresponding to each section. In one embodiment, an exponential function of 2 to the nth power is used as the nonlinear function. In one example, if the first scale is 1, the next is 2, the next is 4, 8, 16, 32. It is further normalized between 0-1. If the leftmost partition is 0, the value when entering the second partition from the left is 1/32, the value when entering the third partition is 1/16, the value when entering the fourth partition is The value when entering the 1 / 8th and 5th sections is 1/4, the value when entering the 6th section is 1/2, and the value when entering the rightmost section is 1. In FIG. 15, when the knob of the slider bar is located in the leftmost section, it corresponds to 0, and when the knob of the slider bar is located in the rightmost section, the position of the knob is non-linear when the knob is located in the middle section. It is factorized. In FIG. 16, when the knob of the slider bar is located in the leftmost section, it is 0, and when the knob is located in the right-hand section, the position of the knob is nonlinearly factorized.

In the non-linear slider of FIG. 15, the leftmost section of the slider bar is “not related at all”, “don't know”, and the rightmost section is “otherwise not buy”, and the position of the knob on the slider bar is The direction toward the left end is “not very important”, and the direction toward the right end of the knob is “very important”. When the position of the knob is in the leftmost section, the importance coefficient is zero. When the position of the knob is in the rightmost section, the importance coefficient is 1. When the knob is located in the section between the left end and the right end, the position of the knob is non-linearly factorized. In FIG. 15, a slider and a graph are shown, but only the slider is displayed on the display unit of the client terminal. The slider in FIG. 15 is applied in the case of Yes / No specifications, and is set so that the value when “not related at all” is 0, and the value when “otherwise not buy” is 1. For example, if you select “Not related at all” for the attributes “waterproof function”, “camera shake correction function”, and “move with dry batteries”, it will be 0. 1 is selected. In other words, it corresponds to 0.0: not related at all, 0.1 to 0.9: somewhat related, but various degrees, 1.0: otherwise not adopted. Therefore, even in the case of Yes / No spec, an analog answer can be obtained. That is, do you want it? How much do you want with a digital answer (YES / NO)? You can get both answers with one slider for an analog answer (somewhat you want, not much).

FIG. 16 shows the case of the continuous spec. The value in the case of “not related at all” is set to 0, and the value is set to increase as the importance increases. In the non-linear slider of FIG. 16, the left end section of the slider bar is “not related at all” or “don't know”, and the direction of the knob on the slider bar toward the left end is “not so important” The direction toward the right end is “very important”. When the position of the knob is in the “left end” section, the importance coefficient is zero. When the knob is located in the section between the left end and the right end, the position of the knob is non-linearly factorized. In FIG. 16, a slider and a graph are shown, but only the slider is displayed on the display unit of the client terminal. The slider in FIG. 16 is applied in the case of continuous specs, and the value of “no relation” is 0, and the coefficient is nonlinearly corresponding to the position of the knob as it goes to the right side of the slider bar. . For example, for the attributes “large number of pixels” and “light weight”, selecting “not related at all” is 0, and the position of the knob is close to 1 as it approaches the right end.

によって、測定する。
ここで、ｍ：属性のインデックス、
ｆ_ｍ：各商品の属性ｍの指数、
ｇ_ｍ：仮想商品（消費者の求める商品）の属性ｍの指数、
Ｗ_ｍ：属性ｍの重み、
である。
距離はユーグリット距離に限定されるものではなく、例えば、マハラノビス距離を用いても良い。 [F] Distance between the product desired by the consumer and each product The distance between the product desired by the consumer and each product is calculated. Assuming that the consumer wants the “virtual product consisting of the highest weighted indexing level for each attribute”, using the difference between the virtual product and the weighted indexing level of each product for each attribute, The distance between the virtual product and each product is calculated. The “highest value” in the “virtual product composed of the highest value of the weighted indexing level of each attribute” is, in one aspect, the weighted indexing level of each attribute of each product in the product database after narrowing down. It is the highest value. In another aspect, it is the highest value of the weighted indexing level of each attribute of each product in the product database before narrowing down. Alternatively, when the level is indexed between 0 and 1, the value of “1 × weighting coefficient” may be the highest value.
In one aspect, the distance is a weighted Eugrid distance (see FIG. 14),

To measure.
Where m is the attribute index,
f _m : index of attribute m of each product,
g _m: index of the attribute m of virtual goods (goods that consumers want),
W _m : weight of attribute m,
It is.
The distance is not limited to the Eugrid distance, and for example, the Mahalanobis distance may be used.

In a preferred embodiment, the distance is converted into a fit rate. If the distance is small, the fit rate is large, and if the distance is large, the fit rate is small. A method for calculating the fit rate will be described. Fit rate is (% -max for each attribute level) ² square root of the sum of all attributes (that is, the same as the Euclidean distance) 100%, (% of the product level-min)% of the square root of the sum of all attributes of ² Say. That is, in a certain product, if all level values are max, the fit rate is 100%, and if all level values are min, the fit rate is 0%. However, the means for calculating the fit rate is not limited to this, and it may be calculated by the ratio of the sum of the squares so that the difference is large without taking the square root.

In addition, the distance for each attribute is stored in the database as well as the overall distance or fit rate between the virtual product desired by the consumer and each product. By doing so, the divergence situation (deviation attribute) can be presented to the consumer.

[G] Display of recommended products One or more products are displayed as recommended products in ascending order of the calculated distance. Preferably, along with the recommended product, the fit rate and the attribute that is deviated are displayed on a display unit of a terminal (a client terminal or a stand-alone terminal on the network).

The recommendation system according to the present invention will be described taking a recommendation of a digital camera as an example. FIG. 17 shows a product database, in which product attributes and levels are stored for 42 types of digital cameras. Product attributes include “Manufacturer name”, “Product name”, “SLR”, “Medium”, “Small”, “Card slim”, “Deformation”, “Price”, “Number of effective pixels”, “LCD” Includes “Monitor”, “Image Stabilizer”, “Optical Zoom”, “Battery Life”, “Thickness”, and “Weight”. When the level of the product attribute has the attribute or not, “Yes” is 1 and “No” is 0.

FIG. 18 is obtained by normalizing each level using the formula (1) in the product database of FIG. If the attribute level is “Yes” is 1 and “No” is 0, it remains as it is. In the case of price, the maximum value is 0 and the minimum value is 1.

FIG. 19 shows an optimized product database after further optimizing the normalized level in the product database of FIG. The “effective pixel number”, “liquid crystal monitor”, “optical zoom”, and “battery life” are optimized based on the formula (2). The “actual price”, “thickness”, and “main body weight” are optimized based on the formula (3).

FIG. 20 shows a product database after optimization. Here, narrowing down by manufacturers B and C is performed based on information input from the client terminal.

FIG. 21 is a weighted product database. The weighting is obtained by coefficientizing the importance for the attribute input from the client terminal.
In FIG. 21, the weighting coefficient of “price” is 1.00, the weighting coefficient of “effective pixels” is 0.60, the weighting coefficient of “liquid crystal monitor” is 0.00, and the weighting coefficient of “camera shake correction function” is 1.00, “optical zoom” weighting factor 0.40, “battery life” weighting factor 0.50, “thickness” weighting factor 0.20, “body weight” weighting factor 0. 20. Each weighting factor for each attribute is multiplied by the index of the database of FIG.

FIG. 22 shows the calculation of the distance between the highest value of each attribute level and the value of each product level using the weighted product database. The total distance of each attribute is the distance as a whole. For example, the highest value is 0.92 in the prevailing price of the weighted product database in FIG. The actual price of commodity No. 7 is 0.52, and the distance is (0.92-0.52) ² = 0.16. Similarly, the distance between levels in each attribute is calculated. Here, without taking the square root, the total distance between the levels of each attribute is the total distance. The distance of product No. 7 is 1.16 (0.16 + 1.00). Simultaneously with the distance calculation, the fit rate is also calculated. In FIG. 21, the total of all attributes (max-min of each attribute level) ² is 1.58. The fit rate of the product No. 7 is {(1.58-1.16) /1.58} × 100 = 26.5.

FIG. 23 is a database obtained by rearranging FIG. 22 in ascending order of distance (in order of increasing fit rate). Also, the divergence items are shaded. In one aspect, a threshold is provided for the distance between the levels of each attribute, and the distance greater than the set threshold is shaded.

FIG. 24 shows a first-stage refinement screen displayed on the display unit of the client terminal (which may be a display unit of a stand-alone computer). This drawing is displayed on the display unit by, for example, selecting a digital camera on the display screen. One of “single-lens reflex type”, “medium / small type”, “card type”, and “deformation type” is selected. In FIG. 24, “Small / Small Type” is selected from the input means of the client terminal. By clicking the start button from the input means, the narrowing information is transmitted to the recommendation engine and the next screen is displayed. It is sent to the client terminal and displayed. In the recommendation engine, “medium and small types” are extracted from the product database.

FIG. 25 is a narrowing-down screen based on detailed specifications. In response to the question “How much can I pay more than this?”, “20,000 yen”, “30,000 yen”, “40,000 yen” One is selected from “50,000 yen”, “70,000 yen”, “100,000 yen”, and “any amount”. In FIG. 25, “40,000 yen” is selected from the input means of the client terminal, and by clicking the “next” button from the input means, the narrowing information is transmitted from the client terminal to the recommendation engine. At the same time, the next screen is transmitted to the client terminal and displayed. In the recommendation engine, “40,000 yen or more” is further extracted from the extracted product database.

FIG. 26 further shows a narrow-down screen based on detailed specifications, in which one or more desired manufacturer names are selected. FIG. 26 shows a case where Company D is selected. By clicking the “next” button from the input means, the narrowing information is transmitted from the client terminal to the recommendation engine, and the next screen is displayed on the client terminal. Sent to and displayed. In the recommendation engine, those manufactured by D company are further extracted from the extracted product database.

FIG. 27 further shows a narrowing-down screen based on detailed specifications, which allows one type of desired recording medium to be selected. FIG. 27 shows a case where smart media is selected. By clicking the “next” button from the input means, narrowing information is transmitted from the client terminal to the recommendation engine, and the next screen is displayed on the client terminal. Sent to and displayed. In the recommendation engine, a product whose storage medium is “smart media” is further extracted from the extracted product database.

FIG. 28 is a weighting input screen using a non-linear slider. Here, "Battery can be used", "Hand shake correction function", "There are colors other than black and silver", "Price is a little cheap", "Battery is good" Each attribute of “good image quality” is weighted by moving the non-linear slider by the input means of the client terminal. By selecting all weights and clicking the “Next” button from the input means, attribute importance information is sent from the client terminal to the recommendation engine, and the next screen is sent to the client terminal. Displayed.

FIG. 29 is a weighting input screen using a non-linear slider. Here, “High optical zoom magnification”, “Light as much as possible”, “Not too light and not too heavy, moderate weight”, “High shutter speed”, “Good close-up function” , Each attribute of “the liquid crystal screen is large” is weighted by moving the non-linear slider by the input means of the client terminal. By selecting all weights and clicking the “Next” button from the input means, attribute importance information is sent from the client terminal to the recommendation engine, and the next screen is sent to the client terminal. Displayed.

In the narrowed product database, the importance of the attribute is integrated as a weighting coefficient at each level of each attribute.

FIG. 30 shows a recommendation result screen. The recommended products are displayed in ascending order of the overall distance. The fit rate is also displayed. In one preferred embodiment, the divergence status is also displayed. The divergence status is displayed, for example, as a bulleted item (fit rate 85%; non-perfect item-(1) resolution (2) weight (3) price). . From among a plurality of recommended products, it is possible to select a product to be purchased with the divergence status as a clue.

It is a figure which shows the recommendation system which concerns on this invention. It is a figure which shows the flow of recommendation. It is a figure which shows the normalization of data and the nonlinear conversion of data. It is a figure which shows optimization of the dynamic range of data, and nonlinear conversion of data (when there is a peak). It is a figure which shows the nonlinear conversion of data. It is a figure which shows the nonlinear conversion of data. It is a figure which shows the nonlinear conversion of data. It is a figure which shows the definition of the goods using an attribute and a level. It is a figure which shows comprehensive evaluation of the goods by a consumer. It is a figure explaining the change of the value of each attribute by a consumer. It is a figure which shows the gap between an ideal product and an actual product. It is a figure explaining consumer's "buying" when it deviates from a required characteristic. It is a figure which shows the relationship between a specification and the degree of purchase of the consumer in individual evaluation. It is a figure explaining distance calculation of the goods which a consumer asks, and each actual goods. A slider bar and a knob are shown as means for inputting the importance of the attribute. A slider bar and a knob are shown as means for inputting the importance of the attribute. It is a figure which shows a goods database. It is a figure which shows the normalized goods database. It is a figure which shows the product database after optimization. It is a figure which shows the product database after refinement | purification optimization. It is a figure which shows the product database after weighting. It is a figure which shows distance and a fitting rate measurement. It is a figure which shows what rearranged the goods in order of recommendation after distance measurement. It is the screen for narrowing down displayed on the display part of the client terminal. It is the screen for narrowing down displayed on the display part of the client terminal. It is the screen for narrowing down displayed on the display part of the client terminal. It is the screen for narrowing down displayed on the display part of the client terminal. It is an attribute importance input screen displayed on the display unit of the client terminal. This is the attribute importance input screen displayed on the display of the client terminal It is a recommendation screen displayed on the display part of the client terminal.

Claims (18)

A system for recommending one or more products from a group of products consisting of multiple competing products,
A product database that stores the attributes of each product and the level indexed to indicate the degree of influence on the purchase;
Means to input the importance for each attribute;
Means to factorize the input importance,
Means for multiplying the indexed level by the factorized importance to obtain a weighted index level;
Assuming that the virtual product that consists of the highest value of the weighted indexing level of each attribute is the product that the consumer wants, use the difference between the virtual product and the weighted indexing level of each product in each attribute, A means for calculating the distance between the product and each product;
Means for displaying one or more products as recommended products in ascending order of the calculated distance;
Having a recommendation system.   The recommendation system according to claim 1, wherein the display unit further displays a deviating attribute together with the recommended product. The recommendation system is:
The recommendation system according to claim 1, further comprising means for indexing the level of each product so as to indicate the degree of influence on the purchase.   The recommendation system according to claim 3, wherein the indexing means includes means for normalizing a level.   5. The recommendation system according to claim 4, wherein the indexing means includes means for optimizing the normalized level.   The recommendation system according to claim 5, wherein the optimization means includes a normalized level of non-linear transformation.   The recommendation system according to claim 5, wherein the optimization means includes optimization of a dynamic level of a normalized level. The input means includes a slider bar displayed on the display unit, a knob provided movably on the slider bar, and means for selecting a position of the knob on the slider bar. The recommendation system according to claim 1, wherein a coefficient of importance is determined between 0 and 1 corresponding to the position of the knob. 9. The slider bar is divided into a plurality of sections, and a coefficient is assigned to each section, and the position of the knob on the slider bar is converted into a coefficient assigned to the section to which the slider bar belongs. Recommendation system. The recommendation system according to claim 9, wherein the coefficient is set such that one partition corresponds to 0 and the other partition corresponds to 1, and the partition between them is nonlinearly converted to a coefficient. The recommendation system according to claim 9, wherein the section at one end is coefficientd so as to correspond to 0, and the other section is nonlinearly converted into a coefficient so that the coefficient increases toward the other end.   The recommendation system according to any one of claims 1 to 11, further comprising means for inputting a narrowing condition by setting a part of the attribute or / and a level of the partial attribute as a narrowing condition for the product.   The recommendation system according to claim 12, wherein the virtual product is configured with a maximum value in a weighted indexing level of each attribute of the product after narrowing down. In order to recommend one or more products from a group of products consisting of multiple competing products,
A product database that stores the attributes of each product and the level indexed to indicate the degree of influence on the purchase;
Means to input the importance for each attribute;
Means to factorize the input importance,
Means for multiplying the indexed level by the factorized importance to obtain a weighted index level;
Assuming that the virtual product that consists of the highest value of the weighted indexing level of each attribute is the product that the consumer wants, use the difference between the virtual product and the weighted indexing level of each product in each attribute, A means for calculating the distance between the product and each product;
Means for displaying one or more products as recommended products in ascending order of the calculated distance;
Computer program to make it function.   The computer program according to claim 14, wherein the display unit further displays a deviating attribute together with the recommended product. A method of recommending one or a plurality of products from a group of products consisting of a plurality of competing products,
Indexing the level of attributes of each product to indicate the degree of influence on purchases,
Entering the importance for each attribute;
Factorizing the input importance, and
Multiplying the indexed level by the factor to obtain a weighted indexing level;
Assuming that the virtual product that consists of the highest value in the weighted indexing level of each attribute is the product that the consumer wants, use the weighted indexing level of the virtual product and each product to determine the distance between the virtual product and each product. A calculating step;
Displaying one or more products as recommended products in ascending order of the calculated distance;
A recommendation method.   The recommendation method according to claim 16, wherein the displaying step further displays a deviating attribute together with the recommended product.   18. The method according to any one of claims 16 and 17, further comprising a step of inputting a narrowing condition by setting a part of the attribute or / and a level of the partial attribute as a narrowing condition for the product before the step of inputting the importance. Recommendation method described.

Images