Enabling Exploratory Search with Dhiti

Last August, I wrote “Exploring Nuggetize“, in which I described an interface that Dhiti co-founder Bharath Mohan developed to surface “nuggets” from a site and reduce the user’s cost of exploring a document collection. As an experiment, I’m now including a Dhiti widget here on The Noisy Channel. If you look at any single post in a browser, you’ll see a widget at the end of the post (before the comments) that attempts to use the post as a starting point for further exploration.

Please use the comments here to provide feedback. Bharath is eager to improve his product, and I’m eager to improve the experience for all of you!

By Daniel Tunkelang

High-Class Consultant.

8 replies on “Enabling Exploratory Search with Dhiti”

Readers can also use our widget to mark “points of interest” while reading an article. We’ll track these as interests, and tailor the recommendations accordingly.

Eager to get feedback from the community here. It matters a lot to us.


Definitely interesting. And a better way of browsing through related material than an one-dimensional list of related posts. Like the related concepts laid out in this fashion rather than disturbing the overall experience of the blog.


We observe the following patterns across the sites in which we have our widgets:

1) The widget does contribute to increased engagement. We see about 5-10% of readers “interact” with the widget, either to click through on an article, or pivot on a concept or a topic. About 60% of the interactions are clicks on articles.

2) We notice that there’s a higher probability of readers reading the articles fully than normal. This is because of “informed navigation”. By the time they click, they know what the article is going to talk about – so there’s lesser click-baiting.

3) We observe search referrals interact a lot more with the widget. This is because their intent is clear, and often unmet in the article, and they’d like to explore the site for related content. They are also most likely visiting the site for the first time. So there is more likelihood for exploration.

4) When a search query brings traffic to a page, we add it as an “interest” by default. Users tend to pivot the highest on these, which indicates that their intent is still unquenched, and they want to explore the site more for the same query!

5) Through the pivots, the publisher gets to know what their readers to interested to explore around. For example, on The Noisy Channel, in the two days its been installed, we observed that readers pivoted on Information Science, Information Retrieval, Natural Language Processing, Interface Design, Enterprise search engine, Deep Web, HCIR, Twitter, etc. These are often very different from what a publisher already knows from search referrals. Eg: TheNoisyChannel may be in the top search result page for HCIR, but Deep Web, Twitter, Interface Design – are all relevant to readers here, in a different context.

6) The pivots also provide cues to the publisher to create reference pages (like wikipedia) for some topics. Eg: Daniel has written excellent content over time on HCIR, or on Deep Web. But they are spread out across several pages. A search visitor, or someone exploring that topic ideally is looking for a reference page.


Historically my users average about 1.4 actions and 4 minutes per visit. I’ll keep an eye om those numbers to see if they increase, since that would be the clearest sign of increased engagement due to the widget.


Unfortunately my numbers are highly skewed by the thousands of people who have been come to my blog in search of Quora invites. And I don’t get raw data with my free Clicky license — even if I did, I’m not about to invest in breaking it down.

So I’ll have to wait until that wave subsides before being able to do meaningful analysis and finding out if Dhiti is helping. But I’m pretty sure it isn’t hurting, and that’s good enough for me to keep the experiment going.


Comments are closed.