Introduction

There is an ocean of information out there. The repositories of information on the Internet are growing larger and more comprehensive each day. Somewhere out there is the specific information that you are interested in. How can you find it efficiently?

There are Internet search engines, but these are only starting points. The Internet is so vast that a simple search on a word or phrase can yield thousands or millions of articles to check. Most search engines base their ranking on the number of times a word appears, positions of the word, number of links to the site, and so on. You still have to wade through the results to find what is relevant to you. Opening and examining the web pages returned by a search engine to see if they really hold something interesting can be time consuming.

If you are looking for something extremely specific, search engines can be great, but if you are looking for items of more general interest, it can be a problem, since you can only search on one or a few phrases at a time. For this sort of thing, people tend to go to web sites that cover news, a hobby, or a profession, but such sites usually cover a small subset of the Internet, and what if you are looking for articles on a combination of interests? 

What Myallo Is

Myallo is an assistant that acts as an autonomous learning agent, doing searches on your behalf. It uses an interest profile to search for articles that are interesting to you. Instead of giving Myallo one or two words or phrases, you can enter a whole series of topics, and tell Myallo how they are related, and specify exactly how interesting each topic is to you.

Myallo uses one or more search engines when it begins, but these are only starting points. Myallo reads and scans through the result pages themselves, evaluating each one in light of the interest profile. It then predicts your interest in each page and prioritizes the results, displaying the most interesting ones. It also scans the links on the best pages and may follow some of them to see if they lead to more interesting pages.

This takes searching to a new level. Myallo doesn't just find pages that match your search string, it also ranks the pages using the interest profile. Since a profile can describing many and varied topics, specifying how interesting (or uninteresting) each topic is, and how the topics relate, it can make an accurate and personalized prediction of relevance.

In Myallo's search results, it shows the predicted interest for each page it found. If you adjust this to indicate your actual level of interest (by moving a simple slider) Myallo will learn from this feedback. It uses advanced neural network techniques to adjust the interest profile. If you consistently tweak a profile in this way, Myallo should steadily improve its search results.

What Myallo Does

Myallo does three major things:

Organizing Your Interests

Myallo lets you record your interest in a various of topics and specific articles.

In Myallo, an Interest is an object that refers to either a general topic, or a specific article. An Interest holds a name, an "interest level," and some other items.

A Topic is a word or phrase that you specify. It is also the Interest name, and is sometimes used as a search string to find articles. You can specify more detailed search terms if you like.

An Article is a source of text whose location can be specified with a URL. An article is usually a page on the Internet's World Wide Web, though it can also be a text or HTML file on your own computer, or on other computers on your local network.

A detail is a setting inside an Interest that helps Myallo's searching or learning processes.

An Interest Profile is a document you can save on your hard disk that holds a collection of Interests. The Interests in the profile are arranged into an outline format.

Below is a simple Interest Profile. There is an interest in the topic "gardening" and two more interests in the topics "flower" and "vegetable". These two interests are set up as sub-interests under "gardening". They indicate your interest in flower gardening and vegetable gardening.

Interest Profile Window

The sliders indicate interest levels. Gardening is set to the default center position, indicating a neutral, or average, level of interest, while flower has a higher level, making articles on flower gardening attractive, and vegetable has a lower interest, making articles on vegetable gardening unattractive.

  

Finding Interesting Articles

Myallo will use the interests in a profile to help it search for candidate articles that might be interesting. To find candidates that are be associated with your interest categories, Myallo uses one or more "search sites", which are usually Internet search engines. In the above example, it might call upon search sites to find candidate web pages that match "gardening" and "flower".

A search site might find hundreds or thousands of potential candidates. Myallo uses several strategies to narrow down the results into a smaller number of candidate articles. Then, each candidate page is read and evaluated according to the profile to predict an interest level for that page. The Agent selects the pages it feels will be most interesting, and places the results in a list at the top of the profile. Opening any of the result in the list will cause your Internet browser to show you the corresponding article.

The following figure shows a profile with the results of a search:

 

Improving Accuracy

As you browse though the results, you can see the level of interest Myallo predicted. After you view the article, you can adjust the prediction by moving the slider. Myallo uses this feedback to adjust the details in the profile, which will help its future evaluations more accurate.

The following figure shows the original profile after the slider for the "gardening" topic has been dragged toward the right, indicating greater interest. Note that as a consequence, Myallo has automatically adjusted the two sub-topics to indicate slightly greater interest as well: