A: We have harnessed vast amounts of human patent intelligence stored in public databases to serve the whole IPR field. Put simply, IPRally is a virtual patent professional learning from real-world professionals with the latest machine learning technologies.
A: Each year, tens of thousands of patent attorneys use millions of hours for defining inventions in natural language (drafting and prosecuting patent applications). At the same time, thousands of patent examiners in e.g. the USPTO and EPO spend millions of hours for examining the outputs of the attorneys (finding novelty bars for the inventions). In this process, an enormous amount of data is formed that we can learn from to assist both in defining and examining upcoming inventions.
A: One significant thing is the graph approach. We re-arrange all technical information in natural language documents into tree-like structures, which the AI algorithms are trained to digest. These data structures are much more sensible and efficient for a computer to understand and process and at the same time intuitive for a human reader.
Another thing is that to our knowledge our search engine is the only one that utilises real patent examination data to train the AI to find most relevant results. In fact, "utilise" is too mild a word, since the search is heavily based on this data.
These together bring a whole new dimension to the search: the technical relations between features become relevant.
A: Keyword searches are excellent when you know beforehand what you are looking for or have the expertise and time to make complex search strategies with synonyms and Boolean logic and so on. And the time to crawl through hundreds of documents manually. In our approach, computer understanding of synonymity, for example is built-in. Also more weight is put on the technical relationships between features, that is, the overall technical concept which is of interest.
A: Definitely not. The user can visually see how the computer understands the invention / technology. The computer can also explain its own thinking by pointing those passages of documents that have caught its attention. This all is made possible by the graphs and makes a true difference.
A: Our AI can actually learn the logic of patenting. By logic we mean the core concepts of novelty and inventive step. In addition to that, we can really take into account very detailed technical information of patents, which is not possible in rough similarity searches, for example.
A: We honestly do not know yet. In just a short time, we've been able to develop a "search monster" that can do just a few seconds much of the work that takes hours or days using traditional methods. Every month, the search accuracy is increased as the algorithms and quality of training data improve.
A: In many ways. First, the user will be presented with technically more relevant search hits and is able to make quicker and better IPR decisions, be it in protecting own inventions or coping with competitors' patents. She also saves time as the computer is able to do much of the raw analysis of the texts and show only the crucial passages.
A: For some companies, immaterial property is their key asset. For some others, patents are just one of the inevitable boring things that need to be taken care of in order to stay in the competition. We want to serve them all, bringing not only a productivity boost but also a pinch of joy to the traditionally tedious IPR work. Our design principle from day one has been to keep the product as simple as possible so that an every engineer or researcher, whoever develops or innovates, can use it, yet powerful enough for patent professionals to achieve their goals quicker.
A: The most active development areas in our approach currently are accurate graphification of the patent documents, domain-specific embedding of patent terminology and tuning the learning capability of neural network algorithms to the maximum. In all these areas, we have had several breakthroughs during the past months and have a lot of development ideas for the future. We will also constantly be adding new patent data and new and higher-quality training data. In addition, there are quite a few related product ideas that just wait to see the daylight. And last but not least, we will listen to our users wishes and implement features that serve as many customers as possible.
A: There is quite a lot of hype around AI these days. We feel and know that our approach is in the very heart of machine learning and AI is not just a buzzword. Solving a difficult problem in a complex domain with real-world data is something that inspires us and attracts a lot of interest around. We also want to make clear that the machine learning and NLP technology is moving fast and the journey has just begun. The real transformation of the way patents are being made, found and evaluated is still to be seen.
A: We like to say I-P-Rally. But we do not mind if you call us your IPR ally, since that's what we really want to be. Rally stands both for speed and change. And speed of change too. Things that people do not typically associate with the patent process but we see inevitable.