{"id":294077,"date":"2016-09-19T00:38:57","date_gmt":"2016-09-19T07:38:57","guid":{"rendered":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/?post_type=msr-project&p=294077"},"modified":"2025-03-04T10:47:26","modified_gmt":"2025-03-04T18:47:26","slug":"enterprise-dictionary","status":"publish","type":"msr-project","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/project\/enterprise-dictionary\/","title":{"rendered":"Enterprise Dictionary"},"content":{"rendered":"

<\/a>1.\u00a0\u00a0\u00a0 Project Introduction<\/h2>\n

“Everyday we are faced with a sea of acronyms, ever changing group structures, and fast-tracked projects.”<\/p>\n

Currently, collation and curation of corporate knowledge is a painstaking manual process. We seek to move these activities into the background so that the relationships between different people, project updates, and emerging milestones can be surfaced in an ambient light-weight way.<\/p>\n

This is our project: Enterprise Dictionary<\/strong>. It is a research project which aims to learn enterprise entities and their properties automatically from enterprise documents. It can gracefully collect information about workplace collaborations, project ebb-and-flows so that institutional knowledge can be easily gathered and shared.<\/p>\n

The ultimate goal of our project is to build an Enterprise Satori.<\/strong> So that we can enable others to build knowledge-empowered applications in Enterprise.<\/p>\n

 <\/p>\n

<\/a>2.\u00a0\u00a0\u00a0 Framework<\/h2>\n

\"Enterprise<\/p>\n

Our framework includes the offline<\/strong> part and the online<\/strong> part.<\/p>\n

<\/a>3.\u00a0\u00a0\u00a0 Offline Part<\/h2>\n

<\/a>1)\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Data Source<\/h3>\n

Currently, we can handle different data sources, such as PDF, WORD, PPT, ONENOTE, MEETING, EMAIL, SharePoint\/Web Page, etc.<\/p>\n

<\/a>2)\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Data Parser & Cleaner<\/h3>\n

We parse all data sources to two formats: XML<\/strong> and TXT<\/strong>.<\/p>\n

XML<\/strong> is semi-structured. It can be used for semi-structured knowledge learning. E.g. Using the Kable\u2019s technique to extract people-project relations.<\/p>\n

TXT<\/strong> is unstructured. It can be used for acronym\/expansion mining, definition mining, project-concept (type) mining, etc.<\/p>\n

<\/a>3)\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 People-Project Mining<\/h3>\n

We apply Kable (semi-structure mining) technology to extract facts from the documents.<\/p>\n

Actually, the real technology is more complex than table extraction. We need to identify the “structure” first, then extract the “data”, followed by classify the content to it’s “function\/property”, and then, build the association. This technology is successfully applied in Bing DU to automatic extract knowledge for Satori.<\/p>\n

<\/a>4)\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0 Definition Mining<\/h3>\n

We start from Hearst patterns, isA patterns, and some other patterns, e.g. “is a”, “is one of”, “means”, “stands for”, \u00a0etc. to get the candidate definition sentences for each topic. We manually label some ground-truth for the definition, and use Rank\u00a0SVM to train a linear classifier to classify whether it is definition or not. The features includes: NLP features, pattern features, data source, author, acronym, structure, statistical, symbol, conceptualization, embedding, keywords, etc.<\/p>\n

Our discriminative features<\/em><\/strong>:<\/p>\n