ARES Analysis

ARES Analysis is nonprofit NLP research project developing information extraction system that process, analyzes and aggregates news from BBC and Reuters on daily basis. It extracts triples in the form: subject(noun) - relation(verb) - object (noun). For example sentence "Trump meets Putin" is extracted as Trump(subject) -> meets (verb relation) -> Putin (object). The software is being written in Java.

Extraction is based on chunking (Shallow Parsing), so resulting model must be improved in the future with Deep Parsing approach. Shallow Parsing processes text as flat structure (not tree structure that is domain of Deep Parsing). Regarding extraction data, you can choose between chunks and superChunks. Chunks are less acurate then superChunks in regard to searching terms but contain more context data then superChunks.

ARES Analysis gives you opportunity to perform search on news data in different ways. For example you can search for every place,person,event etc. (object) that U.S. President Donald Trump visits/visited/will visit if you type subject:Trump verb relation: visit. Or you can search for all interactions/verb relations between EU and China if you type subject: EU object:China. If you want to perform just simple full text search on all news sentences in database, just type something into "Sentence" field.

If you want to use logical operators in search boxes (subject, verb relation, object, sentence) you can use AND OR operators. For example: Trump OR Putin visit OR meet Merkel AND Macron. Each field can contain either AND or OR operator but not both.

ARES Analysis uses combination of custom ARES algorithm and Recursive Neural Network for sentiment analysis of sentences. It means you can search for example for all data that contains verb relation "invest" and have positive sentiment. Please be aware that sentiment data are not 100% accurate because our sentiment analysis model is still in the process of training.

If you want to have a look on overall statistics of some entity, let's say "Japan", just type the keyword into "Entity statistics" field and you will get statistics based on subject, verb relation, object and sentiment data.

Ares Frequency Relations Matrix: displays N-ary relations with frequencies among entities in adjacency matrix (graph theory)

Ares Relations Aggregation: displays aggregated binary relations between entities

Last extraction algorithm runs:

BBC --> processed sentences: 157, processed URLs: 166, finished time: 2020-08-11-09-33-38
Reuters --> processed sentences: 1303, processed URLs: 312, finished time: 2020-08-11-09-44-14
Subject Verb relation Object Sentence Select extraction mode:
Select sentiment Select table Date (MM-DD)
Entity statistics