JP2011070422A - Goods recommending device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a goods recommending device which can recommend goods for a user, even if there is no buying history or browsing history of the user and can recommend appropriate goods to the user. <P>SOLUTION: Keywords for each quiz question or goods information are set, in advance, and goods document vectors related to the keywords of the goods information are prepared, as well. Then, if an answer is given to the quiz question, it is determined whether the answer is right or wrong. A correct-answer document vector is generated on a question with a correct answer, on the basis of the keyword and a wrong-answer rate (a), and a wrong-answer document vector is generated for a question with a wrong answer, on the basis of the keyword and a correct-answer ratio (b). The degrees of similarity between the goods document vector and the correct-answer document vector and between the goods document vector and the wrong-answer document vector are obtained, respectively, and the goods information having a high degree of similarity is extracted and is presented to the user. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a technique for recommending a product to a user via a computer network.

  Conventionally, products have been recommended to users via a computer network. At this time, in order to recommend a product that the user is interested in as much as possible, a recommended product is determined using a past purchase history (see Patent Document 1), or a product that meets the user's search condition is selected. (See Patent Document 2).

Japanese Laid-Open Patent Publication No. 2001-265808 Japanese Laid-Open Patent Publication No. 2007-41971

  However, the technique described in Patent Document 1 has a problem that the purchase history, browsing history, and the like of the user must be collected in advance. Further, in the technique described in Patent Document 2, the product information cannot be used when it does not have the attribute information to be searched, and the user directly inputs the search key. There is a problem that it is difficult to appeal to the needs. In addition, as one of the potential needs of users, there is a viewpoint of satisfying the desire for knowledge by re-learning the knowledge of the weak field while learning the knowledge of the specialty field. If a user has specific knowledge but is unknown to many other users, that knowledge is knowledge of the user's area of expertise and the user does not have specific knowledge Is known knowledge for many other users, it can be said that the knowledge is in the field that is not good for the user. Recommending products related to knowledge in the field of expertise is preferable for users because it leads to further acquisition of knowledge in the field of expertise, and recommending products related to knowledge in the field of weakness Since it leads to acquisition, it can be said that it is preferable for the user.

  Therefore, the present invention makes it possible to recommend products even if there is no purchase history or browsing history, and in particular, the user himself / herself is not aware without explicitly entering a search key. It is an object of the present invention to provide a product recommendation device capable of recommending an accurate product to a user after determining potential needs.

  In order to solve the above problems, in the present invention, a user terminal and a processing server are connected via a network, and according to an answer from the user terminal to the quiz problem, an apparatus recommending a product to an answerer, The processing server stores quiz information in which keywords for a quiz question are recorded, answer history information for each quiz question, and product information in which product contents and product document vectors related to keywords attached to the product are associated with each other. Each quiz is based on the determination means by the storage means, the correctness determination means for determining whether the answer result received from the user terminal is correct, the answer history information of the quiz problem corresponding to the answer result, and the determination result by the correctness determination means. An answer document vector creating means for calculating a score of each keyword attached to the question and creating an answer document vector of the answerer; Similarity calculation means for calculating the similarity between the created answer document vector and the product document vector stored in the product information for each product, and the product information for the calculated product with high similarity There is provided a product recommendation device having product presentation means for extracting and presenting to the user terminal.

  According to the present invention, a keyword is set for each quiz question and product information, and a document vector related to the keyword is obtained for product information, and when the answer to the quiz question is made, the correct answer rate is calculated. Since the document vector of the keyword according to the correct answer / incorrect answer was used and the product with high similarity to the document vector was presented, the product recommendation is made even if there is no purchase history and browsing history It is possible to recommend an appropriate product to the user.

  According to the present invention, it is possible to recommend a product without having a purchase history and a browsing history, and it is possible to recommend an accurate product to a user.

It is a lineblock diagram showing one embodiment of a goods recommendation device concerning the present invention. It is a figure which shows the quiz information 21 memorize | stored in the memory | storage means 20. FIG. It is a figure which shows the merchandise information 22 memorize | stored in the memory | storage means 20. FIG. It is a figure which shows the keyword candidate information 23 memorize | stored in the memory | storage means 20. FIG. It is a figure which shows an example of the document vector (a set of a keyword and a score) about a certain product. It is a figure which shows the mode of the question screen displayed on the user terminal. It is a figure which shows an example of a correct / incorrect determination result and answer log | history information. It is a figure which shows the mode of calculation of the document vector of the problem which was a correct answer, and the problem which was an incorrect answer. It is a figure which shows the presentation screen displayed on the user terminal.

(1. Device configuration)
DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the drawings. FIG. 1 is a configuration diagram showing an embodiment of a product recommendation device according to the present invention. In FIG. 1, the processing server 10, storage means 20, administrator terminal 30, and user terminal 40 are configured.

  The processing server 10 includes a related keyword extraction unit 11, a product document vector creation unit 12, a correct / incorrect determination unit 13, a document vector creation unit 14, a similarity calculation unit 15, and a product presentation unit 16. It is connected so that the data in the storage means 20 can be handled. The related keyword extraction unit 11 and the product document vector creation unit 12 perform pre-processing on the administrator side, and the correctness determination unit 13, the answer document vector creation unit 14, the similarity calculation unit 15, and the product presentation unit 16 Processing is performed on the answer result received from the user terminal 40.

  The related keyword extracting unit 11 extracts keywords from each quiz problem in the quiz information 21 and each product in the product information 22 using the keyword candidate information 23. The product document vector creation means 12 calculates a TFIDF (Term Frequency Inverse Document Frequency) for each product keyword in the product information 22, and creates a product document vector that is a document vector for each product. The correctness determination unit 13 performs correctness determination on the answer result received from the user terminal 40. The answer document vector creation means 14 creates answer document vectors for each answerer, using the received answer results and the answer histories of all answerers, divided into correct answer questions and incorrect answer questions. The similarity calculation means 15 calculates the similarity between the product document vector and the answer document vector. The product presentation means 16 extracts a product document vector having a high degree of similarity and presents the product to the user terminal 40 of the answer source, and uses the function of the WWW server software for the presentation.

  The processing server 10 is connected to the administrator terminal 30 and is connected to the user terminal 40 via a network. The processing server 10 is actually realized by incorporating a dedicated program into a general-purpose computer connected to an external storage device and having a network communication function such as the Internet. In the example of FIG. 1, the processing server 10 and the storage unit 20 are shown separately for the description of the function. However, in reality, the storage unit 20 may be another program processed by the processing server 10, A program for realizing each means in the processing server 10 is stored and configured as a part of the processing server 10.

  The storage means 20 is realized by an external storage device such as a hard disk connected to a computer, and stores quiz information 21, product information 22, keyword candidate information 23, answer history information 24, and document vector information 25. The administrator terminal 30 is a terminal device having an input device such as a keyboard / mouse and a display device such as an LCD, and is realized by a general-purpose computer capable of inputting / outputting data to / from the processing server 10. The user terminal 40 is a terminal device that can transmit and receive data to and from the processing server 10 via a network such as the Internet, and is realized by a general-purpose computer having a communication function.

  FIG. 2 is a diagram showing the quiz information 21 stored in the storage means 20. As shown in FIG. 2, the quiz information 21 includes items of a problem ID, a question sentence, an option, a correct answer, an explanation sentence, a category, and a keyword that specify a problem. As will be described later, keywords are created from question sentences, choices, correct answers, and explanation sentences. In the example of FIG. 2, an initial state before keyword creation is shown.

  FIG. 3 is a diagram showing the product information 22 stored in the storage unit 20. As shown in FIG. 3, the product information 22 includes items of product ID, product name, product content, category, and keyword that specify the product. As will be described later, a keyword is created from a product name and product content. In the example of FIG. 3, an initial state before the keyword creation is shown.

(2. Pre-processing on the administrator side)
Before actually recommending a product to the user, the administrator performs necessary pre-processing using the product recommendation device. First, keyword candidate information 23 is prepared. The keyword candidate information 23 is a list of keyword candidates that are predetermined nouns, as shown in FIG. Since each keyword candidate of the keyword candidate information 23 is for determining the keywords of the quiz information 21 and the product information 22, the choice of the quiz information 21, the correct answer, the noun that is the category of the product information 22, and the quiz information 21 A set of nouns obtained as a result of performing morphological analysis on the product content of the question text, the comment text, and the product information 22 is extracted, and the one suitable for the keyword is determined by human judgment.

  When an instruction is given from the administrator terminal 30 to the processing server 10, the related keyword extraction unit 11 extracts keywords from each quiz problem of the quiz information 21 using the keyword candidate information 23. Specifically, for each keyword candidate in the keyword candidate information 23, a complete match search is performed for each question in the quiz information 21 with respect to the question sentence, choices, correct answers, and explanation sentences. Use keywords. Then, the extracted keyword is recorded in the quiz information 21 in association with the problem ID.

  Subsequently, the related keyword extracting unit 11 extracts keywords from each product in the product information 22 using the keyword candidate information 23. Specifically, with each keyword candidate of the keyword candidate information 23, a complete match search is performed on the product name and product content for each product in the product information 22, and the matched keyword candidate is set as the keyword of the product. Then, the extracted keyword is recorded in the product information 22 in association with the product ID.

  After the keywords are determined for each product in the product information 22, the product document vector creation means 12 calculates TFIDF (Term Frequency Inverse Document Frequency) for each product and its keyword, and each keyword and its keyword for each product. A set of scores (TFIDF values) is created as a document vector of the product. An example of a document vector (a set of a keyword and a score) for a certain product is shown in FIG.

  Specifically, for each product in the product information 22, the number of appearances of the corresponding keyword in the product content is obtained as TF (Term Frequency). Further, the number of occurrences of the keyword in the product content of all products is obtained as DF (Document Frequency). Finally, IDF (Inverse Document Frequency), which is the inverse of DF, is multiplied by TF to obtain TFIDF. Since TF is the number of times that appears in a specific product, the larger the TF, the more important the word of the product is. On the contrary, since DF is the number of times that appears in all products, the larger the DF, Indicates a generic word that is included in any product description. Therefore, the larger TFIDF is, the more important the keyword is for the product. After calculating the TFIDF of each keyword as a score, the score creating means 12 sets a set of keyword scores for each product as a document vector.

(3. Answers to quizzes and product recommendations)
With the score recorded in the product information 22, the processing server 10 publishes the quiz site on the Internet by the function as the WWW server. As a result, the user terminal 40 can access the processing server 10 via the network. A user who accesses the quiz site on the processing server 10 from the user terminal 40 shifts to a quiz question screen by performing a predetermined operation. In the processing server 10, in response to an instruction from the user terminal 40, questioning means (not shown) selects a problem from the quiz information 21 according to a predetermined rule, creates a questioning screen, and sends it to the user terminal 40. Send. The predetermined rule for selecting a problem can be designed as appropriate. For example, when a user selects a quiz category, 10 questions are randomly acquired from the quiz information that matches the category. It can be set as a rule. Here, the state of the question screen displayed on the user terminal 40 is shown in FIG. When the user selects what is considered to be the correct answer from the choices for the displayed quiz question and gives an instruction to transmit the selection result to the processing server 10, the user terminal 40 processes the selection result as the answer result. Send to server 10.

  In the processing server 10, the answer result is received from the user terminal 40, and the correctness determination unit 13 determines whether the answer result is correct. Specifically, for each question, the answer result and the correct answer of the quiz information are compared, and if they match, the answer is determined to be correct, and if they do not match, the answer is determined to be incorrect. Such questions, answers, and correctness determination of quizzes via the network have already been performed as well-known techniques, and details thereof are omitted. As shown in FIG. 7A, the correct / incorrect determination result by the correct / incorrect determination means 13 is correct (indicated by “◯” in the figure) or incorrect (indicated by “x” in the figure) in association with each problem. It is acquired as information about whether it was. Further, the correctness / incorrectness determination means 13 records correct answer and incorrect answer information for each question as answer history information, which is answer history information for all answerers. As shown in FIG. 7B, the answer history information is a record in which the number of questions indicating how many times the questions have been asked in the past and the number of correct answers indicating how many times the questions have been correctly answered are recorded in association with each other. . For example, when a correct / incorrect determination result as shown in FIG. 7A is acquired, 1 is added to the number of questions asked as Q1 to Q10, and Q1, Q4, Q5, Q8, and Q9 are added. One is also added to the number of correct answers to the questions that are presented. The answer history information is used to obtain a correct answer rate (correct answer rate) and an incorrect answer rate (incorrect answer rate) by dividing the number of correct answers by the number of questions.

  When the correctness / incorrectness determination by the correctness / incorrectness determination means 13 is performed, the answer document vector creation means 14 creates an answer document vector that is a document vector for the answer of the answerer. In creating the answer document vector, first, the answer document vector creating means 14 calculates the correct answer rate and the incorrect answer rate of the problem corresponding to the received answer result, based on the number of questions in the answer history information and the number of correct answers. In the answer history information, answer histories for all the questions prepared as the quiz information 21 are recorded, but here, only the correct answer rate and the incorrect answer rate corresponding to the 10 questions that are presented are calculated.

  Two answer document vectors are created: a correct document vector for the correct answer and an incorrect answer document vector for the incorrect answer. Specifically, for correct answers, the keyword weight of each correct question is set to “1”, and the correct answer rate (incorrect answer rate) of the question is multiplied. Add with. The wrong answer rate is obtained by 1-correct answer rate (correct answer rate). As a result, a correct document vector having a score for each keyword is obtained. The correct document vector has a larger value as the error rate is higher. FIG. 8A shows how the correct document vector is calculated.

  On the other hand, for an incorrect answer problem, the keyword weight of each question that is an incorrect answer is set to “1”, the correct answer rate of the problem is multiplied, and all the questions that are incorrect answers are added in units of keywords. As a result, an incorrect answer document vector having a score for each keyword is obtained. The incorrect document vector has a larger value as the correct answer rate is higher. FIG. 8B shows how the incorrect document vector is calculated.

  When the correct document vector and the incorrect document vector are obtained, the similarity calculation unit 15 calculates the similarity between the correct document vector and the incorrect document vector with the product document vector of each product information. As for the correct answer problem, the degree of similarity between the product document vector shown in FIG. 5 and the correct answer document vector shown in FIG. 8A, and as to the incorrect answer problem, the product document vector shown in FIG. 5 and FIG. The similarity of the incorrect answer document vector shown is obtained. Specifically, when the correct or incorrect document vector is d [i] and the product document vector is d [j], the similarity Sim between the correct or incorrect document vector d [i] and the product document vector d [j]. (D [i], d [j]) is calculated by the similarity calculation means 15 executing processing according to the following [Equation 1].

[Formula 1]
Sim (d [i], d [j]) = d [i] · d [j] / (| d [i] || d [j] |)

  As shown in [Formula 1], the similarity Sim (d [i], d [j]) is obtained by calculating the inner product of the document vectors d [i] and d [j], the norm of the document vector d [i], and the document It is obtained by dividing by the product of norms of the vector d [j]. It is desirable to calculate the similarity for all products, but in order to save the calculation, it may be set to execute only for products that match the category of the quiz problem.

  When the similarity between the correct / incorrect document vector and each product document vector is calculated, the product presentation unit 16 extracts product information having a high similarity and presents it to the user terminal 40. Specifically, first, product information with high similarity is extracted for each of the correct answer problem and the incorrect answer problem. High similarity means that the value of the similarity Sim (d [i], d [j]) is large. The number of product information to be extracted is a number set in advance. Therefore, the product presentation means 16 extracts a predetermined number of product information with higher degrees of similarity for each of the correct answer problem and the incorrect answer problem. As described above, the correct answer document vector reflects the error rate of all the answerers. Therefore, when answering a so-called difficult problem, the correct answer document vector value of the answerer becomes large. In addition, since the correct answer rate of all the answerers is reflected in the incorrect answer document vector, if the answer is incorrect to a so-called easy problem, the value of the correct answer document vector of the answerer becomes large. Therefore, a product having a product document vector having a high degree of similarity to the correct document vector is a product related to a problem with a low correct answer rate, and the answerer is a field (known field) where the answerer is good compared to other people. Become. In addition, a product having a product document vector having a high similarity to the incorrect document vector becomes a product related to a problem with a high correct answer rate, and the answerer is a field (an unknown field) in which the answerer is not as good as other people. It will be.

  Subsequently, the product presentation unit 16 creates presentation screen data in which the extracted product information is arranged at a predetermined position. In the presentation screen data, the answer result is arranged, and the extracted product information is arranged separately for the correct answer problem and the incorrect answer problem. Then, the presentation screen data is transmitted to the user terminal 40 that is the transmission source of the answer result. Actually, the presentation screen data is created as HTML data and is transmitted by HTTP.

  The user terminal 40 displays the received presentation screen data on the screen. Specifically, the WWW browser interprets the received HTML data and displays it on the screen. As a result, a presentation screen as shown in FIG. 9 is displayed, and the user receives recommendation of a product suitable for each of the fields known and unknown. In FIG. 9, a rectangle indicated by shading indicates a product image, and actually the appearance of the product is displayed. The example of FIG. 9 shows a case where three known fields and three unknown fields are presented. The recommended product further increases the specialty field and learns the weak field again, so that it is useful for the user and will motivate the user to purchase.

  The preferred embodiments of the present invention have been described above. However, the present invention is not limited to the above embodiments, and various modifications can be made. For example, in the above-described embodiment, the processing server 10 of the product recommendation device includes the related keyword extraction unit 11 and the product document vector creation unit 12, extracts keywords from the quiz information 21 and the product information 22 as preprocessing, and product information The product document vector of each product of 22 is created. This pre-processing is separately created by a computer other than the product recommendation device, and the product information having the quiz information having the keyword, the keyword, and the product document vector from the beginning. May be prepared in the storage means 20.

  The present invention can be used to recommend products to users in the information processing industry using computers.

DESCRIPTION OF SYMBOLS 10 ... Processing server 11 ... Related keyword extraction means 12 ... Product document vector creation means 13 ... Correct / incorrect judgment means 14 ... Answer document vector creation means 15 ... Similarity calculation means 16 ... -Product presentation means 20 ... Storage means 21 ... Quiz information 22 ... Product information 23 ... Keyword candidate information 24 ... Answer history information 25 ... Document vector information 30 ... Administrator terminal 40: User terminal

Claims (5)

A user terminal and a processing server are connected via a network, and in accordance with an answer from the user terminal to the quiz problem, a device that recommends a product to the answerer,
The processing server
Storage means for storing quiz information in which keywords for a quiz question are recorded, answer history information for each quiz question, and product information in which product content and product document vectors related to keywords attached to the product are associated;
Correctness determination means for determining correctness of the answer result received from the user terminal;
Based on the answer history information of the quiz question corresponding to the answer result and the determination result by the correctness determination means, the score of each keyword attached to each quiz question is calculated, and the answer document vector of the answerer is created Answer document vector creation means;
Similarity calculation means for calculating the similarity between the created answer document vector and the product document vector stored in the product information for each product;
Product presentation means for extracting the product information and presenting it on the user terminal for the calculated product with high similarity;
A product recommendation device comprising: A server computer connected to a user terminal via a network and recommending a product to an answerer according to an answer from the user terminal to the quiz question,
Storage means for storing quiz information in which keywords for a quiz question are recorded, answer history information for each quiz question, and product information in which product content and product document vectors related to keywords attached to the product are associated;
Correctness determination means for determining correctness of the answer result received from the user terminal;
Based on the answer history information of the quiz question corresponding to the answer result and the determination result by the correctness determination means, the score of each keyword attached to each quiz question is calculated, and the answer document vector of the answerer is created Answer document vector creation means;
Similarity calculation means for calculating the similarity between the created answer document vector and the product document vector stored in the product information for each product;
Product presentation means for extracting the product information and presenting it on the user terminal for the calculated product with high similarity;
A processing server comprising: The answer document vector creation means calculates a correct answer rate or an incorrect answer rate for each question based on the answer history information, and multiplies each keyword by the correct answer rate or the incorrect answer rate to answer the correct answer document vector or the incorrect answer document vector. Create as a document vector,
The similarity calculation means calculates a similarity between the correct document vector or the incorrect document vector and the product document vector;
The product presenting means presents a product having a high similarity to the correct document vector as a product in a field of expertise or a product having a high similarity to the incorrect document vector as a product in a field of weakness. Item 3. A processing server according to item 2. The storage means further stores keyword candidate information that is a list of keyword candidates,
Using the keyword candidate information, each quiz problem of the quiz information, related keyword extraction means for extracting a keyword from each product of the product information,
A product document vector creating means for calculating a keyword score of each product in the product information and creating a product document vector that is a document vector for each product;
The processing server according to claim 2, further comprising:   The program for functioning a computer as a processing server in any one of Claims 2-4.

Images