Informative Snippets in Results for Efficient Relevance Judgement during the Prior Art Search.
“An attempt to help people separate the wheat from the chaff efficiently.”
Traditional Prior Art Search | Recursive & Time-Taking
Prior art search is a recursive process. You begin by:
- articulating a technical idea in the form of a query,
- feed it to a search engine,
- wait for it to spit out the results
- and then you go through the results one by one.
The relevance of results generally drops as you go down the list. So you refine your query to steer towards more relevant results, and the process repeats.
As you would know, only a few of these results are relevant, and the rest are irrelevant, often termed “noise.” It is not uncommon to see a 50:1 noise-to-relevance ratio in your results.
Going through hundreds of documents to find one relevant piece of information takes most of your time. Mostly you are just spending time reading documents that you will eventually discard. Therefore, judging a document’s irrelevance is generally beneficial as soon as possible. Unfortunately, most search tools put less emphasis on this part. Either they don’t help you evaluate the relevancy quickly or go only as far as showing some highlighted keywords. But these approaches are seldom sufficient to inform you about the document’s relevance.
With PQAI, however, we are putting a lot of emphasis on this. We believe that enabling searchers to judge the relevance of documents quickly is one of the most impactful areas where prior art search engines need improvement.
In this article, let us analyze this problem in some detail, ponder upon possible solutions, and see how PQAI aims to be of help.
Notorious Titles & Irrelevant Text
Let’s assume we are doing a prior art search through patent literature to keep things simple. Unfortunately, patent titles are notorious for being vague and non-informative. See this patent, for example – The title of this patent is – “Method,” that’s it. I admit this is an extreme example, but judging patents by their titles is generally challenging. Even abstracts, more often than not, are difficult to understand. In fact, abstracts may not even relate directly to your query when you are running a search through claims/descriptions. For many results, therefore, you have to judge the relevance by opening those documents in a separate tab and then going through the complete text, trying to pin down the relevant sections, if any.
A typical patent contains about 10-12 pages of text. However, we routinely bump into patents that are longer and have 50-60 pages of text! When looking for prior art, the information you’re looking for could be anywhere within that text. Even expert searchers spend 90% of their time searching for that crucial piece within the text.
How PQAI helps in Time-Efficient Prior Art Search?
When PQAI identifies results, it goes one step beyond. It also picks out relevant parts of the documents matching your query. We call these “snippets” or “passages” – they are complete sentences or select parts of sentences that make sense on their own. They allow you to judge the relevance of a result directly from the search results page. Thus, you spend much less time sifting through irrelevant results is reduced. Of course, you may still need to read lengthy documents, but only the relevant ones. The ones for which snippets aren’t sufficiently informative, but overall, the number is greatly reduced.
The figure below shows you what PQAI snippets look like.
“A head mounted device” can be described as “an apparatus that fits on a user’s head.” Or a “housing positioned in front of eyes” can be described as “device that covers the eyes like a set of goggles“
A Quick Recap
While searching for prior art, you need to spend a lot of time reviewing irrelevant results versus the relevant ones. Getting an idea of the irrelevance of the search result without having to read it all can help you save a lot of time. To reduce this time, PQAI brings the query element mapping feature into it.
When you search with PQAI, each result is accompanied by a query mapping table. The first column shows a part of the invention query, and the second column shows the relevant text from the search result. This mapping is not just word-to-word, but it’s highly contextual. Example: “A head mounted device” is intelligently mapped with “an apparatus that fits on a user’s head.”
Now, doesn’t that sound interesting? So do give PQAI a try. We look forward to hearing about your experience.