Automate processing of unstructured data from google search results

I am Eduardo Schnadower, a PhD student at Carnegie Mellon University Heinz College. The goal of the project is to try to determine whether sponsored search results offer better prices than organic results when using Google to search for products. We already scraped 140,788 search results, but processing them is taking a lot of time, as we are currently doing it with humans. What we need your help for is to use ML to either reduce or completely eliminate the need of human processing. Here it is what the humans are currently doing: A) They are shown a screenshot of the rendered search result, along with a reference image and text describing the product. They also have the option to see a downloaded HTML of each search result, but because it is saved locally, the rendering may not be good. B) Based on the screenshot and/or HTML, the human tells us: a. The price of the product (this was already done with a tool called DiffBot, but we have the human confirming as the results extracted by Diffbot, as they may not be 100% accurate). b. Whether it is the same model as shown in the reference image and text description (for example, if the reference description of a laptop says 4gb of ram and the result shows 8gb, it’s not a match. If a stroller’s description or photo says it’s blue and the result shows a green one, even if it is otherwise identical, it is not a mach. If the result is offering an additional bundled item that is not mentioned in the reference text or image, it is not a match. c. Whether the product is available for sale and in stock. d. Whether the product is sold in US Dollars or another currency. But this has been taking a very long time and we are still very far from the end. We think that someone with expertise in machine learning can help us considerably reduce the amount of work that needs to be done by humans.