Each semantic key acts as a category for a set of keywords. For example, the semantic key “colour” could refer to any number of colours. According to Lexxe CEO Dr Hong Liang Qiao, the ability to use semantic keys is invaluable when users don’t particularly know what terms they should be looking for.
Qiao provided the example of searching for what colours Ferraris are available in. In his case, not all colours are known and searching for “Ferrari colour” doesn’t return results that the user would be looking for.
“For Google, it’s just a matching of the word colour, or maybe colours, and then Ferrari. If you happen to see red or black near these two words, it’s just a coincidence because Google does not know what black really means — it’s part of a colour — but we know this. Our search will be translated into ‘Go and find red or yellow or white or blue, et cetera, plus Ferrari’.”
In Qiao’s example, Lexxe provided a snippet of statistics that show what the most common colours of Ferraris are and their percentages. Similarly, replacing the colour semantic key with price results in statistics can allow the user to get a feel for how much one might cost.
The idea may sound similar to Wolfram Alpha‘s concept of computing statistics, especially since both Wolfram Alpha and Lexxe use natural language processing. However, according to Qiao, Wolfram Alpha is really a computational database, with limited sets of information.
“The weakness for Wolfram is, for all the information you have already manually inputted into your database, once matched, it is very accurate. However, if it is not there, it is not.”
Qiao showed several examples in which Wolfram Alpha was unable to provide a result, including the speed of a kangaroo, turtle or a chicken.
However, it is still early days for Lexxe, which currently is in beta. At the moment, it only indexes 300 million pages using its servers in Silicon Valley. It plans to increase that to 1 billion this time next year, 3 to 4 billion the year after, 10 billion after that and then 20 billion.
Also, while it has 500 semantic keys already defined, it is aiming to expand this to 10,000 in the next 24 months.
The keys also need to be created by someone to help define them. Qiao hopes that as more people use the search engine, they will contribute to help defining the keys in a similar manner to how people contribute to Wikipedia. Qiao expects this to be especially important since the keys will inevitably cover areas of expertise that are too broad for him and his team to work through.
Iin addition to expanding, Qiao hopes to increase the features the search engine can offer. At the moment, the company is experimenting with sentiment analysis by analysing results to see how users feel about a particular topic.
Qiao said that the team had been working with other companies on sentiment analysis and that, during last year’s federal election, it had used Lexxe to correctly predict the outcome.
“We also do airlines’ sentiment analysis. Virgin Blue came up on the top, followed by Qantas and then Jetstar.”
The feature is expected to be included when Lexxe comes out of beta this time next year.