Microsoft Search: Search your document like you search the web
Greetings, Insiders! We are Derik Stenerson from the Word team and Maria Kang from the Bing Search & AI team. We’re excited to share an early look at how we’re using Microsoft Search to make searching within your document easier and more natural. We’re introducing improvements to the Find feature in Word, a great example of Microsoft AI at Scale powering next generation AI experiences – in this case, Office.
How often have you dug through a document to find something because it didn’t use the exact words you searched for? We know from your feedback, this happens frequently, and you’ve come to expect more. We regularly hear, “why can’t this work like search on the web? Why can’t I ask questions and not worry about misspellings?”
We’re happy to share an insider’s look at how we’re addressing this feedback starting in Word on the web and coming soon to all platforms.
Microsoft Search: Search like the web
We’re utilizing well-established web search technologies, such as query and document understanding, and adding deep learning based natural language models. This allows us to handle a much broader set of search queries beyond “exact match.”
- Typos: When there is a misspelling in the query, search can now show related matches. For example, technincian vs. technician.
- Forms of words: When there are different forms of the word in document and query. For example: tech, technology, technologies”; or USA, U.S.A, United States, United States of America; or newborn, new born, new-born, etc.
- Synonyms: For example: citation, quotation, quote, reference might be all the candidates for a term that you are searching for inside the document.
- A multi-word query: A single word query might lead to too many search results. A modern semantic search can often yield better results. For example: as a query of Oil Price, the content having Price of Oil, Prices of the Oil, Cost of Oil can offer related matches from within the document content.
Get answers with Q&A
With the recent breakthroughs in deep learning techniques, you can now go beyond the common search term-based queries. The result is answers to your questions based on the document content. This opens a whole new way of finding knowledge. When you’re looking at a water quality report, you can answer questions like “where does the city water originate from? How to reduce the amount of lead in water?”
An Insider’s Look at the Technology
All of this is made possible using Microsoft Turing model for natural language representation, a large-scale AI model bringing the power of deep learning to a search box near you.
Microsoft has developed two deep learning models that are using the public Bing web data to enable SmartFind feature:
- Encoder model: It outputs vectors that capture the semantic meaning of a text, such that semantically similar text have similar numerical values, to better understand the language.
- Question answering model: It is built using the state-of-the-art large pre-trained Turing language model, the same model that powers question answering and captions in Bing Search.
These updates are rolling out to Word on the web to organizations and users who are opted in to Targeted release in English. It’ll be available in Production soon.
If you have any feedback or suggestions, you can submit them using the Help Improve Office button in the top right corner or click Help > Feedback. You can also post your feedback here. Learn more about what other information you should include in your feedback to ensure it is actionable and reaches the right people. We’re excited to hear from you!
With the Office Insider newsletter, you can get the latest information about Insider features in your inbox once a month! Sign up here.
Comments are closed.