Patent offices face increasing volumes and complexity of applications that necessitate sophisticated prior art searches. AI solutions powered by highly structured data can dramatically increase search efficiency and results to deliver improvements in examiner productivity, workflows, timeliness, and customer service.
Recently, CAS collaborated with INPI Brazil on AI-enabled workflow optimization to accelerate chemical searches and decrease examination times, leading to shorter examination periods overall. Similar improvements are now evident at other patent offices globally.
Artificial Intelligence (AI) is the technology that allows computers to mimic human learning and performance, with wide application in fields as diverse as healthcare, image and video processing, natural language processing, robotics, wireless spectrum monitoring and more. AI also forms an essential part of autonomous vehicles, virtual assistants and many other devices and systems.
AI can significantly shorten the time required for data-heavy tasks like processing loan applications or detecting fraud, while legal services have also increasingly adopted AI for document review and translation purposes. AI allows attorneys to quickly understand complex documents more quickly while spending more time analyzing each case and negotiating terms.
AI can simplify patent prior art searches by quickly and efficiently locating relevant patents while eliminating duplicates, as well as by helping identify any potential infringements or risks that might occur during application processing. AI may also assist in assessing whether an invention meets all the legal criteria necessary for patentability such as novelty.
AI should not be seen as a replacement for human searchers. While AI cannot fully replace them yet, it can serve as an efficient way of reducing work involved with conducting searches and improving results quality. For its success and safety reasons, its implementation must be approached carefully and methodically.
It’s critical that systems can be trained to detect and respond to context in ways that benefit users, such as highlighting relevant results or suggesting expansion queries based on what words appeared most frequently in search results. Furthermore, software should be capable of correcting for errors that arise such as “GIGO (garbage in, garbage out),” where incorrect or unsuitable data enters an algorithm and results in inaccurate or misleading output.
AI will have an enormously profound effect on IP research, but it must be remembered that it cannot replace human searchers entirely. With technology constantly developing and changing, keeping up-to-date is crucial to using the best tools available for your specific needs.
Machine learning (ML) is the technology at the core of most artificial intelligence (AI) applications today. ML algorithms can use large amounts of data to detect patterns that make predictions or decisions and feed back into an ML algorithm for continuous improvement over time. When used for patent searching purposes, machine learning software can identify key terms to focus on to facilitate faster searches; automate repetitive tasks and free up human analysts for more in-depth legal analysis tasks.
AI can interpret and understand natural language to respond to people’s queries or answer them themselves, often seen with chatbots on websites where users interact with machines using natural language. Sometimes this may require human operators supervision in order to interpret or direct AI whereas sometimes AI may rely solely on its algorithm for decision making without human input.
AI relies heavily on deep learning technology, which uses multilayered neural networks to process and analyze data. To train these algorithms effectively, a great deal of data and compute power are required – this makes GPUs essential tools in this line of work.
Many companies employ AI technologies to enhance data integrity, reduce human error and boost efficiency through automation tools such as robotic process automation (RPA). RPA uses software specialized for replacing manual, rule-based processes. Generative AI emerged during this decade – it can produce content such as essays, solutions to problems or realistic fakes of people based on photographs or audio recordings – prompts for which it then produces something unique in response. Some refer to these technologies as augmenting intelligence (AI).
AI can assist patent searchers by automating repetitive tasks and freeing human analysts up for more complex legal analysis. AI tools also assist by highlighting key terms within patents that should be prioritized when reading them as well as finding more relevant prior art references. Some ML patent search tools even score groups of patents and PGPubs according to an indefiniteness score – providing useful measurements of claims’ strength of indefiniteness.
Natural Language Processing
Natural Language Processing (NLP) is an area of artificial intelligence focused on understanding written and spoken text, while NLP technologies help improve and automate the analysis of unstructured data such as emails or transcripts by recognizing patterns and recognizing meaningful entities.
One of the best-known applications of Natural Language Processing (NLP) is translation – but it goes far beyond simple word conversion. NLP algorithms must understand both individual words’ meanings in context with intent as well as how many there are in a source text to ensure accurate results. Obtaining such precision requires tremendous amounts of data analysis and extensive modeling in order to be achieved successfully.
NLP can also be applied in other areas, including sentiment analysis and text generation. NLP is increasingly being employed by businesses as an important element of customer-facing services; helping users find what they are searching for or the answers to their inquiries quickly and efficiently. A company might use NLP to suggest related search terms or provide results using images instead of keywords as examples of using NLP effectively.
Educators are increasingly using NLP to aid student learning. For instance, when asked by a student “Tell me about volcanoes”, an AI chatbot will respond with information regarding those volcanic features. NLP can also be used to identify students who may need additional support and provide extra resources if needed.
The USPTO is making use of artificial intelligence to streamline its patent examination process with tools designed to facilitate prior art search and classification tasks. These tools analyze patent application texts using natural language processing (NLP), machine learning technology (ML), and other AI technologies.
While these tools will not replace human examiners entirely, they can significantly decrease the time it takes for these tasks and enhance the quality of prior art results. With these new tools available to them, USPTO examiners will have more time and capacity for more complex and value-adding patent related work.
Deep learning is one of the most widely utilized forms of machine learning, comprising algorithms that learn from raw data to perform classification, clustering, and pattern recognition tasks. Deep learning technology has proven itself effective for text mining, speech recognition and image analysis applications as well as text mining for text mining purposes. At USPTO we are using deep learning technology as part of patent examination processes in order to help examiners efficiently search prior art more quickly while increasing quality work production while decreasing overall time on each case.
AI based patent search tools present many challenges; as research by the Intellectual Property Office (IPO) and Cardiff University demonstrates, these AI search tools cannot yet replace expert human searchers in patent searching. Instead, these AI search tools currently complement human searchers by performing tasks such as suggesting keywords and query expansions as well as grouping hits together and offering suggestions about which search strategy might be appropriate for use.
Unfortunately, these tasks are complex and require high levels of expertise from human in the loop to understand results, identify problems and make necessary adjustments based on evidence – making this an extremely challenging task for AI systems that is unlikely to be completed quickly.
AI systems often operate behind closed doors with little transparency into how their algorithms work, creating frustration for search requestors when questions about why certain documents were found or not found arise. An experienced searcher can quickly establish why certain documents were or weren’t located before adjusting search parameters to capture more references and increase confidence in results.
The good news is that these challenges are easily surmountable! All it takes to find an AI based search tool that’s user-friendly is finding one with intuitive controls that allow for rapid, professional-grade results and give users flexibility in refining searches, testing iterations, iterating on results, and collaborating with team members. NLPatent provides such an intelligent patent search platform using supervised deep learning AI with user-centric design that immediately sets you on the path toward finding what you’re searching for and allows filtering and refining of search results until finally find what you need! Here is how deep learning can be used in this context
Data collection and indexing
Deep-learning algorithms can be used for the collection and indexing of vast amounts of data from patent databases, scientific journals, and other sources. These algorithms are able to efficiently categorize and organize the data. This makes it easier to find relevant information.
- Identifying data sources: First, determine which data sources are relevant to your application. Data sources for prior art searches and analyses include databases of patents, scientific publications, academic journals, and reports, as well as conference proceedings and technical reports.
- Scraping Web: Scraping Web is a technique that’s used to extract data from online databases and websites. Web scrapers are designed to automatically crawl and collect data from different sources. To avoid legal issues, it is important to adhere to the terms of service as well as copyright regulations for the data sources.
- Datasets and APIs: Certain data sources offer Application Programming Interfaces that provide direct access to the data. This is a structured and controlled method of gathering data. Some organizations also offer datasets that are specifically designed for analysis and research, which can help train AI models.
- Preprocessing and Cleaning: After the data has been collected, the data must be cleaned to remove noise, duplications, irrelevant information, and other inconsistencies which could hinder analysis. Data cleaning is essential to ensure the reliability and quality of a dataset.
- Database creation: Data collected and preprocessed is organized in a structured database. This could be either a relational database or a NoSQL one that is optimized to handle large volumes of unstructured information.
- Extraction of Metadata: Metadata such as title and authors, publication dates, keywords, and patent classifications are extracted from documents and assigned to each record in the database. These metadata are crucial to enabling effective searching and filtering.
- Natural Language Processing Indexing: NLP techniques for textual data are used to create an index that allows fast and accurate retrieval. NLP tools are able to tokenize text, identify entities, such as names and organizations, perform part-of speech tagging, and create word embeddings, which represent the semantic meaning behind words.
- Keyword Indexing and Phrase Indexing Traditional keyword and phrase indexing is also used in addition to NLP based indexing to ensure compatibility with standard search algorithms. This is especially useful when dealing with simple search queries, or working with legacy systems.
- Image indexing: Image indexing is used if the data contains images or diagrams. Deep learning models are used to extract images’ features, and these features can then be used to index visual data.
- Time Series Indexing: When dealing with time series data, an appropriate indexing strategy is employed to optimize queries relating to temporal aspects.
Challenges and Benefits:
- Data collection and indexing allow efficient and scalable searches and analyses of vast amounts information.
- By properly indexing, you can identify prior art and other critical information more quickly.
- Data collection that is automated can be more efficient than manual methods.
- Data collection can be a challenge due to issues such as copyright, data quality and biases.
Semantic similarity and document retrieval
Deep Learning models can be used for calculating semantic similarity of patent claims, descriptions or prior art documents. This can help find relevant prior art documents that might not use the exact wording, but share similar concepts.
Deep learning algorithms can be used to analyze images and diagrams in patent documents. This can help find visual prior art which may not be easy to search using traditional keyword-based methods.
Topic Modeling and Clustering
Deep Learning can cluster similar scientific papers and patents based on the content. This can help identify key themes and technologies that are related to an invention. This is useful during the prior-art analysis process.
Predictions of Patent Quality
Deep Learning models can be trained to predict quality and relevance based on factors such as technical complexity and patent office classifications. This prediction will help patent examiners and inventors evaluate the strength of patent applications.
Automated patent summarization
Deep learning can be used for the generation of concise and accurate summaries. This will make it easier for researchers, to quickly grasp a patent document’s content.
AI can assist patent examiners by suggesting possible prior art on the basis of claims and descriptions in a patent application. This can improve the accuracy and speed of the patent evaluation process.