Agents

Agents are queries of various types that a user saves for future usage. These can be Boolean or basic queries, and can be iteratively refined with examples.

Like categories, you define an agent with training that is similar to the content that you want to find. You can also define content that you do not want.

Users can search using their agents to find documents that match their interests. You can also set up agents to discard results that have already been read, so that users can always view the most recent content, without repeating results.

You, or your end users, can create agents manually, or you can set up profiles to create agents automatically for users as they search and view documents on your system.

TIP: You can use the Users page in the Control section of IDOL Admin to set up and manage agents.

Agents also allow you to set up Communities  of users with similar interests or expertise.

Train Agents

You define the topic with training (similar to category training), which can consist of documents, text, Boolean expressions, and so on. You can also specify negative training, which indicates any items that the user is not interested in. Training can be accumulated over time to reflect changing interests.

Agent retraining builds up a complete picture of what the user wants over time, without using complicated processes to weight older training, or creating new agents.

Use Agents

Agents find information that a user is interested in, and can be used on-demand, or as a way of recommending content to users. For example, you might want to use agents to:

  • recommend articles on a Web site.

  • notify security personnel when a large number of new documents match a security-based agent.

  • implement a wish list on an e-commerce site, and inform users when new products match their interests or wish list items.

  • notify lawyers when new documents that match their latest legal case are made available.

You can also use agents in the alerting process, to e-mail users content that matches their interests.

Conceptual Agents

The most common and versatile type of agents are conceptual agents. These are automatically generated in the process of training using sample text or documents that represent the agent’s topic. For example, to create an agent on ‘tennis’, a dozen documents about tennis could be selected by a user and passed to the agent training process to create that agent.

Categories are conceptual agents, as are user profiles and in general any agent can be used interchangeably for a variety of purposes.

When a conceptual agent is created, IDOL creates a lists of terms and weights to represent the text, documents or topic that is used as training. An example agent, trained on documents about Shakespeare could result in:

SHAKESPEAR~[421] ROMEO~[387] JULIET~[351] HAMLET~[319] MACBETH~[301] OTHELLO~[294] WILLIAM~[224] PLAI~[221] …

In this the numbers represent statistical weighting applied to the terms. The tilde (~) indicates that the terms are already stemmed. The terms and weights can be edited if needed (for example to remove a particular term, or to truncate the list of terms after a certain point) but in most cases, they should be left alone and will be updated automatically as the agent is retrained.

Boolean Agents

Instead of representing topics or concepts, agents can represent Boolean expressions. To match a Boolean agent against documents in an IDOL index simply send a normal query with the Boolean expression as the query text. To store one or more Boolean agents in an IDOL index so that the agents that would match a particular piece of text or a document can be identified, use an AgentBoolean document.

Boolean agents can be combined with conceptual agents by adding a Boolean restriction to a conceptual agent. Or to look at it another way, create an AgentBoolean document with the conceptual agent terms as its DRECONTENT. This will return the agents conceptually closest to a piece of text, but only if they also satisfy the Boolean restriction.