Best practices for Amplified searching
This article is a step-by-step guide to effective and efficient searching in Amplified.
Searching in Amplified starts with an AI search using natural language text or patents to quickly find relevant patents. Next, you'll iterate with a combination of keywords and the relevant results you find to refine and expand.
For this guide, we'll be using a Custom project. Let's get started!
Start by giving context
First, describe what you're looking for
Amplified uses text descriptions and patents to rank patents by overall similarity. You can add text describing what you are looking for or lookup a patent by number.
Next, quickly scan initial results
Take a few seconds to skim the titles in the first page of results. Results should capture the right technical field and generally not include noise. If the results are noisy, that is almost always because the text description is too vague. This is almost always fixed by adding more text.
📝 To get used to how the system works, try adding a few words at a time and updating results to see how they change.
💊 For help on how to write great descriptions, check out our guide and examples that show what kind of descriptions give the best results.
Add keywords as a reading assistant
Now that you know your initial input works the way you expect, the next step is to define what you're most interested in by adding keywords.
Right now we will only use keywords as a reading assistant to help us understand the results and iterate. This will help us decide how to best use keywords as part of our search.
Briefly scan results
Review the first page of results using the checkmark to mark relevant results. You can use the flag for any that you're not sure about yet. The keyword counts, drawings, and relevant text display are helpful for this step.
What if the results are in the right space but leaning the wrong way?
Sometimes you may get results that aren't focused on what you want. In that case, it's helpful to modify your description with what you feel is missing from the current results.
Once the results look generally relevant, it's usually most effective to skip to using Bullseye keywords.
Example:
Imagine you started searching for "a self-driving taxi that uses gesture recognition to automatically detect and pick-up passengers" and wanted to find results about the sensors but the results seem focused more on algorithms. You change your text to "cameras and sensors used in a self-driving taxi for gesture recognition to automatically detect and pick-up passengers" and find that this pushes the results more towards the hardware as you hoped.
Pro tip: Use keyword counts to understand how the AI is working
The keyword counts can be helpful to identify the gap in the AI's understanding of your text. In the example above if you added keywords for "gesture", "taxi", and "sensor" you may find that most results have a non-zero count for "gesture" and "taxi" but 0 for "sensor". This can also help you figure out which keywords are most likely to help as search filters later on.
Once you've finished reviewing the first page, you can automatically mark the rest as Hidden by using use the shortcut next to the Actions button at the top of the page.
Add bullseye keywords
Initially, the goal is to find relevant results so we can afford to be narrow and precise with keywords. In the example above, flux control is the critical term so we've set the search fields to require a match in title, abstract, or claims by unchecking Descriptions.
If that proves too restrictive, we can always check Descriptions and Update results again. Later on this guide will also cover tools that help expand your search for comprehensiveness, such as similar term suggestions and relevant class codes.
Use the relevant patents you found to expand your search
If you found new relevant results, click Update results to learn from them. Once you are no longer finding new relevant results on the first page, click Update results. Amplified will be able to show new results based on the patents you marked as relevant in the last step.
Again, review the first page, mark patents as relevant or flagged, and then mark the rest as hidden until you no longer find new relevant results on the first page.
Iteratively building a search strategy
By this point, you will have learned a lot about the patents in this space and should have a feel for whether there are likely more relevant results to look for and some ideas and how to find them.
Exploring results from different angles
One option is to repeat the above bullseye strategy for different key features. For example, in the quantum computing example above you could make "flux control" and "hamiltonian" highlight only by unchecking all of the checkboxes then add a new keyword "Josephson junction" and Update results.
Adding synonyms and similar terms
If there's a relevant result out there with these initial bullseye keywords, you'll find it quickly with this approach. After reviewing those narrow results, you can easily broaden your search by using Amplified to quickly add synonyms and brainstorm similar terms. Click on the green + icon next to a keyword group to suggest similar terms.
With both approaches above, continue the same pattern before of reviewing the first page of results, marking relevant patents, and then updating results.
Advanced techniques
Some searches are harder than others. Amplified has tools to help with these cases. Here are some additional options to explore:
- Strategically use wildcards or proximity operators in your keywords
- Upload a list of results from a complimentary tool such as CAS/STN
- Add a traditional query using complex Boolean logic
Use AI for a final quality assurance
As a final step, we recommend turning the filters off and updating results. This is a great way to catch any potentially still relevant results that you haven't found yet.
When to stop?
It depends. For some searches, it's enough to just find one or two relevant results. In those cases you can stop once you're happy with what you've found.
For situations where you need to be more thorough, it is a bit more complicated. Searching in Amplified is very similar to the traditional approach to search in the sense that you iteratively build your search strategy. The difference is that Amplified makes finding initial relevant results faster and assists throughout the process to prioritize review by similarity, thereby reducing noise.
Much like the traditional approach, the answer is to stop when you are seeing diminishing returns. As you save relevant results, you'll see the remaining results gradually become noisier. Adding additional search criteria like a filter will stop improving results. That is a strong sign that you've done a thorough search.
In the most extreme cases, you will end up with inputs in Amplified that mirror a traditional search query. You'll have some number of hits that seem defensible as a good indication of a complete search. (If you need a search history, you can export one based on this.)
The good news is that Amplified will have all of your Relevant, Flagged, and Hidden results already. So you'll be able to review this list in priority order and skip all the patents that you've already reviewed before. Once you get comfortable with Amplified, you'll find that often don't need to review the entire hit list since partway through it becomes nothing but noise.