Release 2(v9.0.2)
Part Number A95244-02
 Home |
 Solution Area |
 Contents |
 Index |
| Oracle9iAS Personalization User's Guide Release 2(v9.0.2) Part Number A95244-02 |
|
This glossary explains terms used in the text and terms encountered in discussions related to personalization and data mining.
A graphical user interface that enables you to manage Oracle9iAS Personalization, which includes (1) scheduling the build and deployment of packages and the generation of reports, and (2) managing the creation and use of recommendation engines (RE) and RE Farms.
This type of model is used for all the recommendation methods except cross-sell. It also allows all types of data source as inputs for predicting any of the interest dimensions. See also Cross-Sell Model.
See Recommendation Algorithms.
A group of similar items. A category is an element in a taxonomy; an abstraction for a group of items or categories. In OP, any item or category can belong to one or more other categories. See also Taxonomy.
Category membership specifies how items and categories are related to other categories. For example, an item can have a SUBTREELEAF relation to a category if it is a descendant of that category. Similarly, a category can have a SUBTREENODE or LEVEL relationship with another category. See also Taxonomy.
This type of model is used only in the cross-sell methods. It allows only either navigational or purchasing types of data source for input, and requires that the interest dimension be directly related to the type of input data source.
OP uses data from four types of sources: ratings, purchasing, navigational, and demographic.
The particular demographic attributes of interest to OP are listed below. They are stored in the MTR in the CUSTOMER table/view, which consists of the following fields.
CUSTOMER_ID
NAME
GENDER
AGE
MARITAL_STATUS
PERSONAL_INCOME
HEAD_OF_HOUSEHOLD_FLAG
HOUSEHOLD_INCOME
HOUSE_HOLD SIZE
RENT_OWN_INDICATOR
ATTRIBUTE1 - ATTRIBUTE50: These are generic attributes that can be mapped to any column in the customers' database or can be null. They provide extra flexibility. The first 25 are for string (VARCHAR2) data; the last 25 (26-50) are for numeric data.
The process of transferring the tables that define a model to one or more recommendation engines after the model has been built. A deployment also establishes the necessary connections between the recommendation engine and the MTR.
See Recommendation Engine Farm (RE Farm).
On some e-commerce sites, vendors promote certain products called "hot picks"; the hot picks might, for example, be this week's specials. The hot pick items are grouped into hot pick groups.
The term I-I is encountered in some detailed error messages. It stands for Item-to-Item, and is an obsolete term for what is now discussed under cross-selling. See Cross-Sell Model.
Specifies the interest dimension that items should be ranked against. The interest dimensions supported in OP are rating, purchasing, and navigation.
The MOR is the Oracle database schema that maintains mining metadata defined by the Oracle9iAS Personalization data mining schema and provides for logging in to the data mining system, logging off, and validating users for the MOR and data mining functionality. Provides core data mining algorithm functionality.
The MTR is a schema containing the data used for data mining. It contains all the data necessary to define and build a package. For OP, the MTR has a fixed schema designed to support the building of models that support producing customer/visitor recommendations. The MTR also stores customer data collected through the REAPI.
A model is a set of tables containing all the data necessary to make recommendations. See also Recommendation Algorithms.
Oracle9iAS Personalization.
An object created using the Admin UI that controls model building and deployment. A package specifies the build settings and other attributes that control how models are to be built, as well as the RE Farm to which the models are to be deployed. After the build is complete, it consists of the database connection information and the rules tables that are deployed to the recommendation engine.
The relative degree of individualization desired in OP's recommendations. A high setting produces the most individualized recommendations, those most highly related to the given user profile. A low setting generates recommendations that are the most popular or common for a given user profile. A low setting would yield "best seller" kind of recommendations, whereas a high setting will produce recommendations that may not be appropriate for many people, but the recommendations may be of higher perceived value.
The term P-I is encountered in some detailed error messages. It stands for Person-to-Item, and is an obsolete term for what is now discussed under aggregated models. See Aggregated Model.
All the data collected about a customer from that customer's sessions. Profiles are stored in the MTR or cached in the RE.
The rating scale for OP should be in ascending order of "goodness". That is, create a scale in which a high rated item indicates that the user prefers that item over items with lower ratings.
Oracle9iAS Personalization bases its recommendations on one of two algorithms: Predictive Association Rules and Transactional Naive Bayes:
Predictive Association Rules:
Transactional Naive Bayes:
For fuller descriptions of these algorithms, see Predictive Association Rules and Transactional Naive Bayes, in Appendix A.
The front end of Oracle9iAS Personalization. Via the REAPI (Recommendation Engine Application Programming Interface), RE provides the following services on a Web server associated with the calling Web application:
A collection of Java classes that enable a Web application written in Java to collect and preprocess data used to build OP models and to produce recommendations from OP.
A group of systems with related OP recommendation engines installed. When a package is deployed to an RE Farm, it is deployed to all members of the RE Farm. See also Web Farm.
This is how Oracle9iAS Personalization makes its recommendations:
See Recommendation Engine Farm (RE Farm).
An object created using the Admin UI that controls when models specified by a package are to be built or deployed, or when a report is to be generated.
With reference to applying a predictive model to new data, scoring means assigning a score that reflects the likelihood that a particular record belongs in a certain class. A score is the confidence in a prediction.
Sessions are used to organize user activities. A session corresponds to a set of activities that a user does in "one sitting". Each session is uniquely associated with a user and has a start_time and end_time. All the activities performed by that particular user within the (start_time, end_time) interval are considered to be part of that session.
These terms apply to the host Web application and indicate whether the application does its own session management or not. OP has its own internal session management in both cases. If the application is sessionful, OP maintains the mapping between its internal session and the application's session. If the application is sessionless, OP's session starts at the first activity of the user and ends when the user has been inactive for a pre-specified time period. See also Web Application Session versus OP Session.
Recommendation Engines can be in any of the following states:
An identifier for an Oracle database instance. In OP, it refers to the unique identifier assigned to each system associated with an MOR. Each system attached to an MOR must have a unique identifier specified in its configuration file.
In the OP context, this term refers to the structural organization of items in a company's catalog or site. Typically the catalog and/or the site has a hierarchical structure, with the most general category at the base (for example, "clothing"), and branching to increasingly specific categories (for example, from "clothing" to "shoes" to "sneakers" to "tennis shoes").
Items can belong to more than one category and to different levels of the structure. For example, "tennis shoes" can be a category in "clothiing" and also a category in "sports equipment."
The structure of the OP taxonomy is a DAG (direct acyclic graph), which can contain multiple top-level nodes. The different portions of the taxonomy can be disconnected too. Any node can connect to any other node but there cannot be a path that connects a node's child back to the node itself.
OP also supports multiple taxonomies (different ways of organizing the items). The taxonomy is implemented using a group of tables (they are specified by the customer at installation time):
The user of Oracle9iAS Personalization is a DBA or system administrator or Java programmer, or perhaps all three. Do not confuse this with the user of a Web site that uses OP.
Anyone who visits or logs on to the Web site. There are two kinds of users:
Either type of user is assigned a user ID by the Web application.
Sometimes the Web site user is referred to as the end user, to distinguish this user from the user of OP.
Some Web applications keep track of sessions, which provide an association between the Web server and a Web client. This association persists over multiple connections and/or requests during a given time period.
Sessions are used to maintain state and user identity across multiple page requests. The Web applications maintain session information in different ways, e.g., by using cookies, by URL rewriting, or via hidden variables like HttpSession objects. A Web application is sessionful if it keeps track of sessions, sessionless if it does not.
OP has its own session management. An OP session maps the OP end user's activities during a certain period, i.e., from the first activity until the session is timed out or is closed by the host application. Whether the host Web application is sessionful or sessionless, OP always manages its own session in order to provide better predictions. If the host application is sessionful, the OP session is perfectly mapped to the host application session. If the host application is sessionless, OP tracks the session on its own, which has no effect on the host application.
A Web farm uses two or more servers to host the same site. HTTP requests are usually routed to each server using some appropriate scheme, such as round-robin routing, to distribute load and allow the site to handle more requests in a timely manner.
During a given session, recommendation requests go to a recommendation engine, because information is temporarily stored there before being synchronized back to the MTR. If you have multiple REs, all the information from any one session has to be kept together for that RE.
|
 Copyright © 2001, 2002 Oracle Corporation. All Rights Reserved. |
|
