Today SezWho a universal profile, content discovery, and a sophisticated reputation engine provider, has announced its acquisition of Tejit, a provider of semantic intelligence solutions. The acquisition enables SezWho to provide more precise contextual reputation scores for contributors based on topics of conversation. ReadWriteWeb gives you an in-depth look into SezWho's latest acquisition and how SezWho measures up to the competition.
Tejit CEO Indus Khaitan began developing Tejit in 2007 as a personal project when he became frustrated reading duplicate content from the 1000+ blogs he had bookmarked. Since then, Tejit has expanded its analysis capabilities across millions of blogs. Tejit's semantic-analysis engine uses Natural Language Processing (NLP) and semantic matching technology to identify topics, sentiments and entities present in web content.
According to SezWho CEO, Jitendra Gupta,
The traditional method of content discovery based on the similarity of content is not adequate for connecting conversation across social sites in a meaningful way. A new level of context-sensitive, semantic discovery is required to reflect all the layers of users' participation across the social web, and to track their contributions in a way that is universally relevant both within and across communities.
There's no doubt that the traditional rating system for comments has its flaws. In a post titled "Disqus Clout: Fail!", Phil Glockner of Scribkin points out one of the biggest flaws of comment rating systems using Disqus as an example. In the comments section, Louis Gray sums up the problem nicely:
I would expect it rewards those who comment most frequently, and wouldn't be so much a subjective view.
Instead of replacing your comment system, SezWho aims to augment the conversations. Keeping the aforementioned flaw in mind, SezWho considers two important factors that: distributed conversations and the people behind them. SezWho provides a meta network information around participants and context. The context has information from various platforms to allow data and content to reside within the community. The service captures valuable information about the history and expertise of individual contributors. Community ratings are only a portion of the cumulative rankings for an overall score.
SezWho provided us with a comparison chart to better demonstrate the differences between what SezWho offers versus competitors Disqus and Intense Debate, which we've previously reviewed.
With all that SezWho adds, it can be argued that some of it will amount to more noise for users. While, we've previously used SezWho here on ReadWriteWeb to enhance our community, some of our writers are using the less complicated Disqus platform on their personal blogs. We wonder if the amount of blog coverage has also affected SezWho's userbase compared to Disqus, which has seen tremendous coverage since its launch.
With SezWho, other important issues are being tackled beyond their enhanced reputation system such as keeping track of conversations over a plethora of platforms and enabling a more sophisticated way to discover relevant content. SezWho aims to enhance communities rather than replace them,but can they filter the noise that's add everyday?
Comments
Subscribe to comments for this post OR Subscribe to comments for all Read/WriteWeb posts
Thanks Cordiva, for the good post. Filtering of the noise and rewarding the right behavior is really what the acquisition of Tejit is about. We want to get more and context specific in figuring out the credibility of the content. You know, one person's signal is another person's noise...So its really important to discern the context and provide relevant credibility and links...
Another important element is the realization that the conversations are going to get more and more distributed and as such no one platform is going to host all conversations. As such its really important to have a distributed service that can work with all content types and platforms to provide a rich context across the web.
That is what we are focussed on at SezWho.
Thanks, Jitendra
Posted by: Jitendra | May 28, 2008 10:55 AM
I share the same point of view concerning Disqus and IntenseDebate.
Their main problems are:
- Backward/forward compatibility: as your comments are stored in their walled garden repository, you can forget about them when you want to get rid of their plugin (unless you go through an import/export process which is not guaranteed)
- Rating is totally subjective. If you support Obama and someone says something about Clinton, you will be tempted to lower rate the comment.
I'm curious to see what the mix between SezWho and Tejit will bring... A commentag like?
Posted by: Xavier Damman | May 28, 2008 10:57 AM
I cant speak for IntenseDebate, but I will add a few points on the post and Xavier's comments in relation to Disqus (and the reason I decided to use them).
1. Comment location is both your site and centralized: with their API WP plugin the search engines see the comments on your site, this was a big selling point for me, and I fully understand the concerns where this isn't possible
2. Walled garden Xavier is incorrect: you can export your comments out of Disqus and back into WP if you decide you don't want to use them any more. They don't current have WP to Disqus (they're working on it) however you can opt to keep your existing comment system/ comments and only use Disqus on posts without comments or for new posts, again, a big selling point for me when I signed up
I also don't get the rating system debate, while it's a novelty I don't think a comment system will live or die by the strength of its rating system, least not today.
Posted by: Duncan | May 28, 2008 4:55 PM
SezWho CEO, Jitendra Gupta said...
The traditional method of content discovery based on the similarity of content is not adequate for connecting conversation across social sites in a meaningful way.
Nope, Jitendra Gupta is completely wrong here or either he is hyping & misleading. Similarity/Proximity is still king today and it will be for the next 100 years or so. Technology can only improve over time and Similarity/Proximity methods are no different. Perhaps Jitendra Gupta meant that Similarity/Proximity classification error needs to improve? This means that mis-categorising rate needs to decrease, such as categorising the bag of words as "Obama, President, White House" into a wrong category of "Movie Stars" rather a category of "Politics". Sure, but as I have stated, technology evolves so as the Similarity/Proximity classification error will improve.
Take a look at those numerous algorithms that were presented at the Workshop on Algorithms for Modern Massive Data Sets in 2006 at Stanford (MMDS-2006). Here is a tip, take a look at the Tensor-Based algorithms. Here is a brief description. Tensor algorithms are multidimensional as compared to current ones which are only 2D (2 dimensional or 2 key-metrics). What does this mean? You can only correlate 2 variables. In text search engine, you can only collect data based on the document and the word-frequency in each document, and these are the 2 variables. Document #1 might contain {Obama, President, White, House} and Document #2 might contain {McCain, Soldier, President, White, House}. The respective frequency for word "President" is 1 for both docs, the word "Obama" is 1 for doc #1 and 0 for doc #2, and so forth for other remaining terms/words. In tersor-based algorithms, one can collect data in many dimensions. Lets say, the analyst is interested in analysing which newspaper is saying what about the presidential campaign. So, in this case the data can be collected as 3D (3 dimensions or 3 key-metrics). These are, news-organization, document and the word-frequency. Tensor-based could even go higher than 3. BTW, tensor calculus is not new. Albert Einstein developed his general theory of relativity in 1916 using tensor. It is only recent that mathematicians, statisticians and scientists realized that it could be applied in data analysis, such as search engine and the likes.
Jitendra Gupta can you tell us your benchmark, ie, the precision & recall of retrieval in using Similarity/Proximity methods compared to Semantic methods? I am trying to see if your claim is just purely for marketing purposes or you have done some benchmarking which convinced you that Similarity/Proximity methods are inferior?
Finally, if you want to get full access to papers on MMDS-2006, then these are available at SIAM journal.
Posted by: Falafulu Fisi | May 28, 2008 5:26 PM
hey Falafulu,
thanks for a good comment...Very intersting.
What I meant was that context around people can help a great deal in discerning the relevance...E.g. for figuring out democratic/republican context, just an analysis of content will not help but looking at people based contexts might provide a better answer.
Interesting references though...I'll check them out.
thanks, Jitendra
Posted by: Jitendra | May 29, 2008 12:11 AM