Prologue

The following scenario is intended to illustrate the capabilities of the Mnemonic Information Management System (MIMS). All results are obtained using the current version of our software, with a graphical user interface that we employ for demonstration purposes.

Scenario

For the purposes of this scenario, let's say that a user is interested in all Reuters news stories that are material to Ford Motor Company's core automotive business. Our scenario consists of the following four steps.

Step 1: Keyword Search

When the user searches for news using the keyword "Ford", 117 hits are returned from a collection of only 19,043 documents. These 117 hits include many uninteresting stories about Ford's credit and defense subsidiaries as well as entirely irrelevant stories about automotive suppliers, car dealers, car owners, and even former President Gerald Ford. This screen shot shows the headlines of the first 30 hits returned by the keyword search.

Even worse, the keyword search has missed several important news stories about the domestic automotive market -- including critical labor relations issues -- that are material to Ford's core automotive business but do not contain the keyword "Ford".

Step 2: User Marks Examples

Mnemonic's software allows users to prioritize information according to their unique requirements. Unlike a keyword search engine, our software is ideal for evolving or recurring searches.

To use our software, the user must identify documents that are relevant along with documents considered irrelevant.

Here the user has marked four documents as relevant (marked in blue) and another four documents as irrelevant (marked in red). Note that all documents are assigned an initial relevance score of 0.50 (50%) because the user's relevance model has not yet been trained.

Step 3: System Creates Model

Next, our system create a statistical model of the user's information need, based on the documents marked by the user. The model creation process occurs in the background and does not interfere with the user's work flow.

Step 4: Model Prioritizes Information

After a relevance model is created, it ranks all documents on behalf of the user. The user may retrieve the new ranking at any time.

This screenshot shows the 30 most relevant documents in the new ranking. The documents marked in blue have been labeled by the user, while those in black have not yet been labeled by the user. Notice that the top articles (ie., the articles with the highest relevance scores) are all material to Ford's core automotive business.

This screenshot shows the 30 least relevant documents in the new ranking. Most of these articles are clearly irrelevant to Ford's core automotive business.

To refine the ranking produced by their relevance model, the user simply marks more documents and asks our system to create a new model. Our system is highly accuracy and very easy to use. It can dramatically improve the user's productivity and the quality of their search experience.

Epilogue

Once a relevance model is created, it is stored for future use and can be used to automatically prioritize new information as it becomes available. The user can also continue to refine their relevance model over time, which will further improve its accuracy.

In the preceeding scenario, the user explicitly marked articles according to their information needs. It is also possible to create relevance models based solely on user behavior. For example, a model can be created based on the documents that a user has retrieved or purchased. The resulting model can be used to recommend documents that the user is likely to retrieve or purchase.