What is 'Hybrid Searching' - and how can this help you find patents?

December 1 2017

Ambercite has been developed to apply the power of advanced citation analytics (and our database of 155 million patent citations) to patent searching.

The most common alternative to citation searching is the widely used keyword searching. And this certainly has its place, even though it can miss relevant patent where inconsistent terminology is used - and create a long list of irrelevant patents to search through, with only a small proportion being relevant.

For this reason, our users use both Ambercite and keyword searching in parallel, and for good reasons. But in most cases these are run as separate searches

But maybe the best answer is to use both Ambercite and keyword searching as part of the same search?

Where query patents are not known, this is common practice - a keyword search is required to find one or more starting patents, and these are used as query patents in Ambercite, which will quickly find many more patents.

But what if you already have a good set of query patents?

For example, a patent being invalidated, or the results of an earlier keyword search?

In this case, a combination of Ambercite Ai to find a set of similar patent, and a keyword filter, can be a surprisingly effective approach - which we could call "Hybrid searching".

Ambercite only recently made abstracts available in its results, and every time I run such a hybrid search, I am pleasantly surprised by just how well it works. 

For example, lets take a look at a recently granted Facebook patent, being US20170293364A1, for Gesture-based control system. It claims a method for hand control of a computer, with an emphasis on the fingers in some of the dependent claims. It is currently listed with 6 patent citations

Facebook image.gif

Imagine you were looking for similar patents - you might even be a patent examiner.

How could hybrid searching help you?

You might simply start with a query along the following lines:

facebook patent query.gif


1000 patents sounds a lot to review, but as will be shown below, it does not need to be.

This will return, in total, 874 patents.

So does this mean that we need to review 874 patents?

Not at all. Because this is a hybrid approach, we can make some assumptions. The first assumption could be that relevant patents are likely to include the word 'hand' in the title or abstract. We can highlight this word in the results:

hand highlighted.gif


And use this as the basis of a keyword filter on the title and abstract field

hand keyword filter.gif

This will reduce the patents to be reviewed to 183 patents. 183 sounds a lot, but they are in predicted order of similarity to the query patent - and in the Ambercite review box it need not take long to understand the focus of a patent can be quickly understood, such as the example below:

Ambercite patent review box.gif


But perhaps there is a more subtle approach. Gesture control based on hands is relative well known, but perhaps basing these on finger movement maybe more novel, and this was the emphasis of the dependent claims in the Facebook patent. So maybe instead we use the word 'finger' as the word filter instead

Finger filter.gif


This reduces the number of patents to review to 75 - and a lot of these relevant, for example:

Finger patent found.gif

And like the previous patent, it is both published before the priority date of the Facebook patent application, and has not been previously cited, i.e is an 'unknown' citation.

So a very useful technique indeed - and one that can help in patent litigation as well, as shown here




However a suggestion. In the above example I ran a couple of different keyword filters. I could have also added the keyword filters 'gestures' or 'control'. These were options, but any additional keyword filters run the risk of filtering out relevant patents. So some judgment of the searcher is required - just like any other search - and I would suggest going broad rather narrow in these judgments. 


Could CPC codes also be used to in a hybrid approach?

The short answer is yes - we do list CPC* codes for patents found, and supply a filter for these.

However a word of caution. After running a large number of Ambercite searches, I have seen many sets of relevant results with a wide range of CPC codes - and have come to be quite nervous about using CPC codes to filter out patents. Nonetheless, we do offer this capability for those users who wish to do this

*CPC - for patent families without CPC codes, we list IPC codes instead. However we do not list IPC codes for patents with CPC codes. While there is often a degree of similarity between CPC codes and IPC codes, the CPC system is the newer system which we presume gives better data. 

Mike Lloyd