I have written about Attivio on the AppGap before (see Attivio Tightly Integrates Structured Data and Unstructured Content for a New Approach to Information Access and Attivio on Some Potential Winners in our New Economic World).  Recently, I talked again with Attivio’s CTO, Sid Probstein and their VP of Marketing, MaryAnne Sinville, on some of their latest moves.  We first discussed Traction’s selection of the Attivio Active Intelligence Engine ™ (AIE) to power their information access.

I know Traction well and have great respect for their offering (see for example, Traction Software Releases Team Page 4.1 – Better Enabling the Social Side of Work).  Sid said that Traction chose Attivio because of its single, flexible API, full Java support, multi-language capability, as well as its granular and secure permissioning model. This makes sense as both platforms are Java-based. More importantly, I am familiar with Traction’s strong granular permissioning capability so it would be important to have a search engine that supports this model.

Sid explained that there are two ways that search can support permission granularity, early bound and late bound. With early bound, the search capability handles permission information as content is loaded. This way users only see in search results, what they are allowed to access.  It is the cleanest and most secure way to handle this requirement. However, if can be hard to accomplish and maintain, especially when changes are made to the status of a large number of documents. Attivio is able to keep up with these demands by storing permissions as structured data – in tables – and using their query-side JOIN operator to link security with content that matches the user’s query and permissions.

The other fall back method is called late bound.  This approach does searches without regard to security levels. Then it checks results against a relational database of security levels before releasing the results. This can cause performance issues as it adds a step and brings back far more documents that then have to be narrowed down.  It also adds an extra layer of expense and maintenance requirements. In addition, there can be security leaks through actions such as spell checking. Moreover, certain features become impossible like facet recognition and partitioning, another capability built into Traction that I cover next.

Sid showed me a sample Traction search page powered by Attivio (see above). The results are offered in the middle column. In the left column are aspects of the results grouped by facet (e.g., key words. projects, labels, types, authors).  This is a very useful way to dig into the content by displaying more than just the titles of documents. You can drill down into these facets to find related content such as other works by the same author. You can also select search parameters such as projects, date ranges, and authors to narrow the results. Below is sample return on authors.

Next, we moved to some of the new features and capabilities within Attivio. They have released a Sentiment Module that extracts and measures the sentiment or “attitude” of documents, as well as the entities within documents. The attitude may be the person’s judgment (e.g. ‘positive’ vs. ‘negative’) or emotional tone (e.g. ‘objective’ vs. ‘subjective’).  Sid explained that the sentiment analyzer is a trainable component that can be used with any language.  The tool comes out of the box trained for common Web usage but you can increase the sophistication to fit your requirements.

For many situations document level sentiment is sufficient. However, for multi-topic content such as long documents and emails, you can get sentiment at the entity level. For example, you can find that a message indicated that a blogger liked the product but did not like the dealer.  You need to train the tool to handle multiple entities in a document but the ability to do this comes with the tool. Here is a sample sentiment alert.

You can also use the sentiment analyzer to look at Twitter and other micro-messaging tools. Here you need to first train the sentiment analyzer on the common text short cuts used in these tools but, again, the ability to perform this training comes with the tool.  I think this capability will be very useful for business intelligence and marketing activity. You can use it both inside and outside the enterprise. While the Web applications might seem more obvious, I also think it will be very useful for looking at internal documents.

Attivio also released a new Classification Module that provides an enhanced means to classify documents to a set of categories. For example, in the media industry, information could be classified by news categories (local, world, business) or classified ads (community, housing, for sale), etc.  Previously they had a rules-based classifier. Now Attivio has brought the machine learning technology used in the sentiment analyzer to enable auto-classification of new documents. You first train the tool on a category through examples, and then it will be able to carry on the task with new documents.  Sid mentioned that these new features are also part of the Traction implementation. With Traction, you can use prior manual classifications to train the classifier, saving a lot of time.

These are all good moves for a product that I have been hearing good things about. Sid said that one of the reasons he is excited about the Traction connection is that they will be able to innovate together. I look forward to seeing what comes out of this pairing.