Dolores Labs’ Technology at Work
- Sentiment Detection
-
O’Reilly Media wanted to find the sentiment of thousands of blog comments on a news event. Dolores Labs labeled every comment as “pro” or “con” in under 24 hours.
View Sample
Dolores Labs also used Turkers to determine what sentiment message board posts about stocks expressed (buy, hold or sell) so that investors could use the data to gauge the community’s feeling about a particular stock.
View Sample - Document Classification
-
Scribd, the “YouTube for documents,” needed to classify the thousands of documents their users upload daily. Knowing whether a document should go in “Health,” “Culture,” “Humor” or one of twenty other categories is a tricky task for a computer, but a very easy one for a human. By using Dolores Labs, Scribd was able to classify thousands of documents and use the data to build an automated classifier.
View Sample - Event Classification
-
Zvents helps people search for fun local events of interest to different users. But they have trouble using automatic classification to label events so that users could find the kinds of things they wanted to attend. Dolores Labs collected 100,000 hand labeled judgments from Turkers that Zvents then used to train their machine learning event classifier.
View Sample - Price Extraction
- We extracted prices from popular shopping sites, many of which are impossible to automatically scrape accurately.
View Sample - Search Relevance
- While most search companies employ a staff of annotators who pore over search result after search result to judge each results relevance for a given search query, Dolores Labs is able to get the relevance judgments of thousands of different web-users within a number of days. Determining search relevance this way is not only faster and cheaper, but it also allows you to collect judgments from a wide swath of Internet users.
View Sample - Image Retrieval
- We collected a set of images for products.
View Sample
To see other uses of our technology, check out our blog.