The Web About Blog

We tried to find the best digital camera brand on the market using our system and it turned out that it gives preference to Cannon. The CNET reviews confirm the choice of our system.

We used the WebAbout system to find the opinions about certain car models like Pegeut 207 and vacuum cleaner producers. The system works quite well. As a matter of fact, it works better when there is a specific model provided at the input, however it also returns useful information when only the trademark name is given.

We try to adapt it to scan opinions about people. Language constructions in which people participate are different than the constructions for not animate objects – the semantic differences result in different language constructions

We improved our algorithm and updated the site.
Now our system captures more opinions and it is more precise due to the new syntactic-semantic rules which were incorporated in it a week ago.

It seems we have to stress more strongly on machine learning techniques, in order to learn more lexica and syntactic rules.

If you see the mobile phone comparison chart, now the things changed, it seems that Sony Ericsson W980 leads, followed by Motorola Z10 and Nokia N78.

We are constantly improving the accuracy of our opinion mining system, so today it is more accurate than it was yesterday.

We measured how accurately the system classifies the sentences into positive and negative, we took in random 21 positive and 17 negative sentences for 7 mobile phones. We measured separately the accuracy for positive and negative sentences:

It turned out that for positive sentences the accuracy was 71% and for negative it was 82%

It is still difficult to make conclusions on the basis of these results calculated from a relatively small sample. However, we think that these results are promising, since they are somehow similar to the results of some state-of-the-art opinion mining approaches.

We already performed opinion analysis for more than 25 mobile phone models. What does our automatic analysis show for each model? It shows the attitude of the people towards the phone. It reflects its quality, its usability, users expectations (big expectations, for example, make people to express negative opinions about iPhone 3G) and other things. As emotional the people are, as many opinions we detect.

On the other hand, the producers themselves use positive emotional terms to promote their products, currently this is reflected also in our analysis.
Our rating currently detects the positiveness of the attitude of the people towards a certain product. This is often well correlated with the quality, but not only.

The project is still in its early stage, but the prototype is being used successfully. We are running the opinion extraction system for a set of mobile and smart phones and it finds what the people think about these gadgets. In fact, the results seem to be very good up to this moment! On our demo page we published the pages generated by the system.

The world changes, and also the way people perceive this world. We believe that in the recent years the Internet has been changing not only the way we access information, but also the way we think and the way we look at the things around us.

The purpose of our project is to improve the way people evaluate certain objects and ideas around them. We began with collecting and summarizing opinions about products. Our technology is intended to facilitate the way one gets oriented about the quality of a product he or she intends to buy. This technology can also be used to collect information about people and companies…and we do not intend to stop here…