Here is another in a series of session notes for KM World 2010 and Enterprise Search Summit 2010. I attended the session, Search for Sentiment led by Seth Grimes, Principal Consultant, Alta Plana Corporation. Here is the description:
“Correctly making sense of sentiment, opinion, and mood-subjective, attitudinal information-poses a special challenge for search, first in correctly indexing sentiment, then in understanding searcher intent in order to respond to search queries, and finally in presenting sentiment-search findings. In this session, Grimes will discuss applications and business benefits in areas including customer support and satisfaction, brand and reputation management, and marketing and product management. He will describe the characteristics of sentiment and opinion and outline technical approaches and challenges. He will provide examples of online sentiment analysis via the application of search technologies to attitudinal information found in traditional and social media. Grimes will also sketch a road map of likely future sentiment-search developments.”
Seth noted that sentiment analysis has been getting a lot of attention recently. Seth says his definition of enterprise search is very broad. For example, he considers Google Web search an enterprise tool because it provides a lot of value to enterprises. This makes sense.
He offered several assertions about sentiment. He first noted that when someone says it is a fact, you know it is an opinion. So he wants to be clear that these assertions are opinions. He feels that human communications are subjective and online facts and have business value. He then offer examples of facts and feelings.
He gave some examples of sentiment tools, beginning with: I Feel Fine. It is a consumer Web tool that can be a toy. I have seen it before and agree. Then he moved to a business site that expresses opinions about travel experiences.
Google search offers some sentiment information. This is one reason that sentiment is a search problem., because Google addresses it other search engines will also cover it. Offering sentiment on things like travel is a decision support effort. Decision support needs to go beyond general-purpose search. For example, you need a summary of opinions. Counting term hits does not meet the need. For example, BlogPulse does not give enough information to make decisions. It just shows mentions but you do not know why or what was said.
Seth said that sentiment analysis is the task of identifying positive and negative opinions, emotions, and evaluations. It turns attitudes into data. However, this definition is five years old and we need to go beyond positive or negative and decide what are the emotions involved.
Seth covered some tools he finds useful. The first was Jodange. It provides faceted search and the ability to do summaries. You can interact with the results. You also need to be able to assign positive and negative within components of a content segment such as a tweet. There are often positive and negative views within the same statement, especially comparative ones.
You fall short with only doc-level analysis as there as there can be multiple sentiments in a single document. You fall short with only keyword-based analysis as you need semantic analysis. You fall short with only human only analysis as you need the power of machines to go through the vast amounts of data. You fall short with only machine only analysis as you need the sensitivity of humans. So you need a hybrid solution.
He showed some more examples of tools. SAS Social Media Analysis was the first and it offers social CRM. Clarabridge provides customer experience management. Crimson Hexagon has a fine grained analysis of perceptions. He showed an Crimson Hexagon example from Twitter political comments around the country where you can drill down by state and see the sentiment. Here is an early commentary I wrote (Essentials of Online Opinion Monitoring from Crimson Hexagon).
Seth said that one of the trends in sentiment analysis is to go beyond polarity. There is some academic work that is surfacing in business tools. LiveJournal offers degrees of happy, sad, and angry. This can be useful in call centers.
It seems that this is a niche in progress but some progress has been made.
Very interesting feedback, and glad to see we were mentioned! (Clarabridge). I should make one note, that we have been offering the ability to look at sentiment "intensity" (on an 11 point scale, or "fine grained") since out 4.1 product available as of March 2010. Since we look down to the concept level, (sentences are broken down into concepts), it is very "fine grained". If curious, we do have a trial version of our self service product which you can get to from our home page http://www.clarabridge.com .
Cheers!
marcos
Posted by: marcos sanchez | November 22, 2010 at 10:03 AM
Marcos. Thanks your comment and link. i will try to check it out. Bill
Posted by: bill Ives | November 22, 2010 at 06:44 PM