JP2007183986A - Item recommendation apparatus based on user's relevance, item recommendation method and algorithm thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an item recommendation apparatus for appropriately obtaining a recommended person's preference and highly precisely recommending items from history of small quantity by noticing relevance between users from user's item reading/purchase (use) and the evaluation of purchased (used) items. <P>SOLUTION: The item recommendation apparatus is constituted of: a user ID acquisition part for acquiring user ID as a recommended person; an item browsing part for browsing items to be browsed in accordance with user's operation; an evaluation acquisition part for acquiring the evaluation of a browsed item which is inputted by the user; a history information storage part for storing user history by storing the evaluation as the meta data of the item; a database for storing the user history; an evaluation request part for requesting the evaluation of the read item to the user again; a recommender specification part for specifying a recommender to the recommended person; and a recommended item specification part for specifying an item to which a high evaluation is provided by the recommender generated by the recommender specification part and which is not purchased or used by the recommended person. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

The present invention relates to an electronic commerce and CRM using a computer, a customer management system, and a recommended item specifying method for recommending items (products, contents, applications, services) on the Internet.

Conventional item recommendation methods use classification based on item attributes (see, for example, Patent Document 1. The nature of an item is determined in advance by determining the nature of the item), or depending on user attributes There are those using classification (for example, see Patent Document 2) and those using user similarity in purchased (utilized) items (for example, see Patent Document 3).
Patent Publication 2004-206679 Patent Publication 2006-236341 Patent Publication 2001-229285

By focusing on the relationship between users based on the user's browsing / purchase (use) history of items and evaluation of purchased (use) items, the user's preferences are accurately grasped, and items are recommended with high accuracy from a small history. An item recommendation device is provided.
Conventionally, an item classification method and a user classification method using a history have been proposed for estimating user's preference. However, how appropriate the classification depends on the properties of the item to be handled, and is not guaranteed. According to the present invention, it is possible to appropriately recommend items without requiring classification of items used for specifying recommended items or classification of users.

A user ID acquisition unit for acquiring a user ID as a recommended person,
An item browsing unit that presents to the user items to be browsed in response to user operations;
For an item viewed by the item browsing unit, an evaluation acquisition unit that acquires an evaluation or a state input by the user, and
A history information storage unit that stores the evaluation or state as user history in the database as item metadata;
A database for accumulating user history;
An evaluation request unit that checks the user history in the database and requests the user to evaluate the item viewed again,
A recommender identifying unit that identifies one or more recommenders to the recommended person using the user history stored in the database;
Using the user history stored in the database,
One or more recommenders generated by the recommender specifying means have given high evaluation,
In addition, the recommended item specifying unit specifies one or more items that are not purchased or used by the recommended person. (Figure 1)

Even if the items and users to be handled are classified in advance or a dynamic classification method is applied, whether it is appropriate depends on the properties of the items to be handled. The utility of using the device could not be obtained unless considered. In addition, when it is difficult to classify items to be handled, or when the relevance between items is sparse, it may not be possible to predict what sort of classification will result in high utility.
On the other hand, the present invention can obtain a high utility without any classification of items and users.
The present invention is expected to provide a high utility with a small history when handling any item.
Note that the utility of the present invention increases as the number of users increases. (In the conventional apparatus, if the number of users is equal to or greater than a certain number, the utility does not change even if the number of users is increased.)
The utility of the present invention is higher as the popularity bias (ease of browsing) of the items becomes more significant. (Even if there is a bias in popularity with conventional devices, the utility does not change.)
The utility of the present invention does not increase even if the relationship between items is strong. (In a conventional apparatus, a high utility can be obtained if the relevance between items is strong.) However, there is a possibility that the present invention can also have a high utility due to the popularity bias caused by the high relevance between items.
In the model “1000 items handled, 100 users, no relevance between items, no bias in popularity”, the present invention and the conventional recommendation device (noting the relevance between items, the recommended person selects the item When purchasing (using), a group of users who purchased (used) the same item recommends one or more items most purchased (used) in the past to the recommended person) and purchased the item on its own without relying on the recommendation device A comparison of the utility at the time of (utilization) was made by computer simulation.
However, 20% of the viewed items are purchased (used), and half of them are highly evaluated. Utility is the probability of giving a high rating to a presented item. (Figure 2)
The present invention has the highest utility when browsing 20 to 30 items (1 out of 3 gives a high evaluation), and 2 of the recommendation devices focusing on the relevance between conventional items when searching by oneself. Has double utility. After that, the utility will be lower, but it will be stable and higher than the case of own power.
The conventional apparatus is more effective than the conventional apparatus unless all users view about 760 items on average and purchase 152 items.
Since this is just an example, the number should be different from the actual operation, but the graph outline and relative positional relationship did not change no matter what input parameters were substituted in the simulation.

Electronic commerce dealing with products such as books that are purchased (used) frequently.

Product recommendation using CRM and customer management system

Providing user-friendly content for search engines

First, the state of the item is defined as (FIG. 3).
When the user logs in to the electronic commerce site in FIG. 1, the user ID acquisition unit acquires the user ID and passes the user ID to the history information storage unit.
The history information storage unit receives the user ID, confirms the user's history information from the database, and if there is an item in the state “b”, passes the item ID to the evaluation request unit.
Upon receiving the item ID, the evaluation request unit requests the user to “please evaluate this item”, and the user gives an evaluation of “satisfied” or “dissatisfied” to the item. The evaluation acquisition unit attaches status labels of “0” and “1” to each and transmits it to the history information storage unit together with the item ID.
When the user searches for an item, the item browsing unit presents the product.
The user can use the radio buttons to “Purchase (use)”, “I am interested but do not purchase (use) now”, “I am not interested, so the system will not recommend this item in the future”, “Purchase in the past ( “I was satisfied with this item” and “I was dissatisfied with this item that was purchased (used) in the past”. In FIG. 1, the evaluation acquisition units attach status labels “b”, “a”, “c”, “0”, and “1” to the history information storage unit together with the item ID.
The history information storage unit receives the user ID / state label from the evaluation acquisition unit, refers to the user history information from the database, and updates it.
The recommender accumulation unit receives the user ID and (v0, v1, va, vb, vc, v *) for the state (0, 1, a, b, c, *) attached to the item for each user. Substituting a numerical value (Equation 1), e [u, u_other] is calculated based on this value, a user u_other with recommendation suitability for u is selected, and the user IDs of the recommended person and the recommender are set as the recommended item specifying unit. To pass.
The recommended item specifying unit receives the nominated person ID and the recommender ID from the recommender specifying unit, refers to the history information of the two persons from the database, the recommender is in the state “1” (FIG. 3), and the recommended person is in the state “ One item “a” or “*” (FIG. 3) is selected at random, and the item is presented to the user who is the nominated person.

It is a block diagram which shows the system configuration | structure of the item recommendation apparatus which concerns on one embodiment of this invention. The probability that the user will be satisfied when receiving a recommendation using the present invention, the probability that the user will be satisfied when receiving a recommendation using the existing invention, and the probability that the user will be satisfied when not using the recommendation device An example of a graph comparing the horizontal axis as the number of browsing. As a specific example, the history information storage unit changes the item evaluation and state in the item metadata stored in the database according to this state transition diagram.

Explanation of symbols

1: The user requests the item browsing unit for an item to be viewed.
2: The item browsing unit presents the requested item.
3: The evaluation acquisition unit requests the user to evaluate the item viewed by the user.
4: The evaluation acquisition unit acquires the evaluation and state for the item viewed by the user.
5: The evaluation acquisition unit passes the evaluation and state acquired in 4 to the history information storage unit.
6: The user ID acquisition unit acquires the user ID.
7: The user ID acquisition unit passes the user ID to the history information storage unit.
8: The history information acquisition unit acquires history information from the database with the user ID.
Overwrite the database with new history information that reflects the item metadata acquired at 9: 5 in the user history information.
If the history information acquired at 10: 8 includes the state to be updated and the metadata of the item to be requested for evaluation, it is passed to the evaluation requesting unit.
11: Based on the metadata acquired in 10, the evaluation request unit makes a user a state to be updated or a request for evaluation.
12: The user ID acquisition unit passes the user ID as the recommended person to the recommender specifying unit.
Using the user ID of the recommended person acquired at 13:12, obtain history information other than the recommended person from the database. 14: Pass the recommender specified by the recommender specifying unit to the recommended item specifying unit.
The item which the recommender acquired at 15:14 gave high evaluation and the history information of the recommended person are acquired from the database.
16: Present the recommended item specified by the recommended item specifying unit to the user who is the recommended person.
21: An initial state in which “*” is added to an unviewed item.
22: State transition when the user is not recommended.
23: State transition when the user receives a recommendation.
24: “b” is added when the user purchases (uses) the item.
25: When an unviewed item is browsed, “c” is added when the user does not purchase and indicates that he / she does not intend to purchase (use) in the future. Items marked with “c” are never recommended again.
26: When an unviewed item is browsed, “a” is added when the user does not purchase and indicates that he / she is “interested but currently does not intend to purchase (use)”.
27: “1” is assigned when the evaluation for the purchased (used) item is high.
28: “0” is attached when the evaluation for the purchased (used) item is low.
29: State transition when giving an evaluation to an item purchased (used) by the user in the past without using the item recommendation device.
30: A state in which no item is purchased (used).
31: A state in which an item is purchased (used).
32: A state where the item is not evaluated.
33: A state in which the item is evaluated.

Claims (11)

One or more users who have similar preferences or one or more users who have recommended aptitudes other than the recommended person from the browsing history or purchase (use) history of the item (product, content, application, service) and evaluation of the item A device that recommends items that the user is satisfied with and that the nominee has not viewed or purchased (used),
User ID acquisition means for acquiring a user ID as a recommended person,
Item browsing means for presenting an item to be browsed in response to a user operation;
Evaluation acquisition means for acquiring an evaluation or state input by the user for an item browsed by the item browsing means;
History information storage means for storing the evaluation or status as user history in the database as item metadata;
An evaluation request means for checking the user history in the database and requesting the user to evaluate the viewed item again,
Recommender specifying means for specifying one or more recommenders for the recommended person using the user history stored in the database;
Using the user history stored in the database,
One or more recommenders generated by the recommender specifying means have given high evaluation,
An item recommendation device comprising: a recommended item specifying unit that specifies one or more items that are not purchased (used) by the recommended person. In the item recommendation device,
The history information accumulating means accumulates metadata about the item (item name, status such as whether the user viewed or purchased (used) the item, evaluation given to the item by the user) in the database as history information,
The recommender specifying means refers to the user ID acquired by the user ID acquisition means in the previous term and the metadata stored by the history information storage means,
2. The item recommendation device according to claim 1, wherein the recommended item specifying unit refers to metadata stored by the history information storage unit. The recommender specifying means specifies one or more users who have similar preferences or appropriate presentations to the recommended user from the history information for each user, and the specified recommender purchases (uses) The item recommendation device according to claim 1 or 2, wherein one or more items are identified from the ones with the highest evaluation given later. Items to be viewed are all products, contents, applications and services traded in electronic commerce,
The history information accumulation means is in a state that an item is viewed by the user,
The rating entered by the user after viewing the item,
A state that the user has purchased (used) after browsing the item,
The evaluation entered by the user after purchase (use) for the item is part of the metadata,
2. The item recommendation device according to claim 1, wherein a part or all of the four metadata is stored in a database. Items to be browsed are all products, contents, applications, and services traded by companies that use the customer management system in the customer management system,
The history information accumulation means is in a state that an item is viewed by the user,
The rating entered by the user after viewing the item,
A state that the user has purchased (used) after browsing the item,
The evaluation entered by the user after purchase (use) for the item is part of the metadata,
2. The item recommendation device according to claim 1, wherein a part or all of the four metadata is stored in a database. Items to be browsed are all products, contents, applications, and services provided free of charge on the Internet.
The history information accumulation means is in a state that an item is viewed by the user,
The rating entered by the user after viewing the item,
A state that the user has purchased (used) after browsing the item,
The evaluation entered by the user after purchase (use) for the item is part of the metadata,
2. The item recommendation device according to claim 1, wherein a part or all of the four metadata is stored in a database. Items to be viewed are all products, contents, applications and services traded in electronic commerce,
The item recommendation device according to any one of claims 1 to 6, wherein the metadata includes a part or all of the four metadata according to claim 4. Items to be browsed are all products, contents, applications, and services traded by a company that uses the customer management system in the customer management system, and the metadata is a part or all of the four metadata according to claim 5 The item recommendation device according to any one of claims 1 to 6, characterized by comprising: Items to be browsed are all products, contents, applications, and services provided free of charge on the Internet.
The item recommendation device according to any one of claims 1 to 6, wherein the metadata includes a part or all of the four metadata according to claim 6. Identify one or more users who have similar preferences or one or more users who have recommended aptitudes other than the recommended person from the browsing history or purchase (use) history of the item (product, content, application, service), A method for recommending an item that the user is satisfied with and that the nominated person has not viewed or purchased (used),
A user ID acquisition step of acquiring a user ID as a recommended person;
An item browsing step for presenting an item to be browsed to the user in response to a user operation;
An evaluation acquisition step for acquiring an evaluation or a state input by the user for the item browsed by the item browsing step;
A history information accumulation step for accumulating the evaluation or state as user history in a database as item metadata;
An evaluation request step for checking the user history in the database and requesting the user to evaluate the viewed item again,
A recommender identifying step for identifying one or more recommenders to the recommended person using the user history stored in the database;
Using the user history stored in the database,
A recommended item specifying step of specifying one or more items that the one or more recommenders generated in the recommender specifying step give a high evaluation and the recommended person has not purchased or used. Item recommendation method characterized by the above. Identify one or more users who have similar preferences or one or more users who have recommended aptitudes other than the recommended person from the browsing history or purchase (use) history of the item (product, content, application, service), A program that operates a computer to recommend items that the user is satisfied and that the nominated person has not viewed or purchased (used),
A user ID acquisition step of acquiring a user ID as a recommended person;
An item browsing step for presenting an item to be browsed to the user in response to a user operation;
An evaluation acquisition step for acquiring an evaluation or a state input by the user for the item browsed by the item browsing step;
A history information accumulation step for accumulating the evaluation or state as user history in a database as item metadata;
An evaluation request step for checking the user history in the database and requesting the user to evaluate the viewed item again,
A recommender identifying step for identifying one or more recommenders to the recommended person using the user history stored in the database;
Using the user history stored in the database,
A recommended item specifying step for specifying one or more items that the one or more recommenders generated in the recommender specifying step give a high evaluation and that the recommended person does not purchase or use; A program to make it run.

Images