Google still can't effectively harness user sentiment to shape search results
In the last week of November, the New York Times ran an article that showed how one retailer could use customer complaints and negative reviews to boost its search ranking in Google. Within a couple of days of the article's publication, Google was receiving complaints of its own, but this time it was from the media, who questioned the search engine's ethical responsibility in the matter.
Not even a week later, Google has replied by applying a new search algorithm that detects and downranks online retailers that "provide extremely poor user experience."
"We were horrified to read about Ms. Rodriguez's dreadful experience," Google Fellow Amit Singhal said yesterday. "Even though our initial analysis pointed to this being an edge case and not a widespread problem in our search results, we immediately convened a team that looked carefully at the issue. That team developed an initial algorithmic solution, implemented it, and the solution is already live. I am here to tell you that being bad is, and hopefully will always be, bad for business in Google's search results."
Singhal did not disclose the way Google determines a good retailer from a bad one, but he did point to the unreliability of Sentiment Analysis, a method of tracking the public's collective opinion on a subject based upon the frequency of keywords in news stories, comments, reviews, and social media. It's a technique that is used daily by online marketers and branding specialists, as well as by financial analysts and others.
In theory, sentiment tracking would take negative commentary about a service and translate it into strikes against that service. The problem with this, however, is that the reliability of a source must be established to determine the "weight" of a comment; and similarly, the degree of negativity per keyword must also be weighted. For example, an expert source calling a service "poor" would be weighted differently from an anonymous source calling that same service "abysmal."
Furthermore, controversial topics tend to have more passionate commentary about them.
"If we demoted web pages that have negative comments against them, you might not be able to find information about many elected officials, not to mention a lot of important but controversial concepts," Singhal said. "So far we have not found an effective way to significantly improve search using sentiment analysis."