<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Pouria Fewzee on Medium]]></title>
        <description><![CDATA[Stories by Pouria Fewzee on Medium]]></description>
        <link>https://medium.com/@fewzee?source=rss-b04d55bde0de------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*jD_CKR3agvDR11G0.jpeg</url>
            <title>Stories by Pouria Fewzee on Medium</title>
            <link>https://medium.com/@fewzee?source=rss-b04d55bde0de------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 13:40:44 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@fewzee/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Getting to 99% Accuracy with Entity Recognition]]></title>
            <link>https://medium.com/@fewzee/getting-to-99-accuracy-with-entity-recognition-d2c0d7a8da7e?source=rss-b04d55bde0de------2</link>
            <guid isPermaLink="false">https://medium.com/p/d2c0d7a8da7e</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[nlp]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Pouria Fewzee]]></dc:creator>
            <pubDate>Thu, 23 Feb 2017 14:47:50 GMT</pubDate>
            <atom:updated>2017-02-27T02:36:15.342Z</atom:updated>
            <content:encoded><![CDATA[<p>One of our main technical goals at Zoom.ai is to make sure that our automated assistant understands our users perfectly, as they engage through chat or email. To do so, we need to confirm that our technology recognizes entities and intents very well.</p><p>Entities are pieces of information that give context to the intent of a query. For example, in the sentence “Schedule a meeting with Andre sometime in the last two weeks of March,” the intent of the sentence is to schedule a meeting, and to make the intent more clear, two entities are included: 1) <em>Andre</em>, which is a person’s name, and 2) <em>sometime in the last two weeks of March</em>, which corresponds to a time span.</p><p>It is this dynamic unlimited nature of these entities that made us build our own core NLP entity extraction technology instead of using off-the-shelf solutions.</p><p>In the past one year, we have been carefully learning from our users’ interactions and that has led us to recently launch a new algorithm for entity recognition that allows Zoom.ai to recognize entities with accuracies greater than 97%. We also now support more entity types such as subject, teams, job titles along with our previously known entity types of dates, people, and locations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FLfROQGL4id8yMY8DfLqlw.png" /></figure><p>To obtain such high accuracy, the new algorithm takes account of semantic and syntactic structures of queries, as well as the order in which words appear one after another. Basically, we have put a more balanced emphasis on the use of grammatical structure of sentences by efficiently traversing the dependency graph.</p><p>Furthermore, since the main goal of our natural language processing solutions is to enable our users to interact with Zoom.ai as seamlessly as possible, we’ve made sure that the capabilities of our models go beyond the grammatically correct use of the language.</p><p>Having the accuracy as the primary goal, we have focused secondarily on the extensibility of the entity recognition algorithm. This is important because as we allow our users to get more and more of their tasks done through their automated assistant, we need to recognize more and more entity types. And, the way we’ve made this possible is by relying on the syntactic and semantic structures of the language, and so requiring a minimal training set per entity type.</p><p>Additionally, since we have access to practically infinite amount of training data per language, as we improve on the accuracy of our language models, the accuracy of our entity recognition algorithm will improve with no extra manual effort required.</p><p>What we’ve got coming out next in the lineup of our natural language processing technologies is the inclusion of intent recognition in the process of entity recognition, hence making use of syntactic and semantic structures of the language, and at the same time allowing one traversal of the dependency graph for getting both tasks done.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d2c0d7a8da7e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>