Introduction

SENTRA is Zoral Lab's advanced, enterprise strength, highly granular sentiment & entity relation extraction and transaction system. Fully automated, based on emerging Sentic computing conceptual framework, using a large array of integrated, highly scalable Artificial intelligence (AI), Machine Learning (ML), Natural Language Processing (NLP) and Logic Programming techniques, it is able to generate highly structured time-series transactions, in real-time, containing sentiment analysis as it applies to any set of entity and entity relationships from an unlimited number of unstructured data sources, (e.g. web sites, documents, blogs, on-line news, emails etc.)

What are sentiment transactions?

There are many systems that will analyse sentiment within text. However, whilst this can be useful, it has limited value when using data for commercial or analytical applications. Just to know that an article or blog entry is negative about its subject is not always sufficient to be really useful. This is best illustrated by an example.

Let’s take the statement:

“Microsoft share performance has been disappointing this quarter, but company's flagship operating system Windows 7.0 has been generally well received and is a big improvement”

Is this positive or negative? Well, understandably, it is both. In order to analyse this properly we need to understand the entities being discussed, their relationship and then the sentiment between them. In other words, we need transactions that clearly structure WHO is saying WHAT about WHOM. This is exactly what SENTRA does. This is what makes it different and uniquely powerful. So, to continue our example, in analysing the above statement SENTRA will generate the following transactions:

Notice, SENTRA understood that the word “company's” refers to Microsoft and was able to understand that Windows 7.0 is a product of a company called Microsoft.

Now let’s take the example:

“Samsung states that its new B5029 model is far more flexible than Nokia’s S23”

This is more complex. SENTRA would generate the following:

Further, relationship between products B5029 and S23 is understood and extracted as that of competitive products.

SENTRA even generates transactions when there is no news

In conjunction with aiHit technology, SENTRA monitors millions of company websites and can see when events/items change, e.g. product releases, changes of executives, customer announcements, partner changes etc. As SENTRA’s AI/ML engines understand the nature of each item, wherever they are located on the web site, it is able automatically generate change transactions.

For example:

How many sources can SENTRA monitor?

Unlimited – SENTRA already monitors and maintains a map of the entire WWW. So the number of sources depends entirely on your monitoring requirements. SENTRA is already being used by the PR industry to monitor over 100,000 sources of news, print media and broadcast sources.

What types of entities does SENTRA monitor?

A wide range including, Company, Products, Person, Events, Event Types, Topic, Sentiment

Does SENTRA identify entity relationships?

Yes. For example it will identify that COMPANY X is a CUSTOMER of COMPANY Z, or that PRODUCT A belongs to COMPANY B etc.

How frequently can SENTRA update its data?

As often as you require. SENTRA can monitor sources in near real-time. SENTRA is available 24/7 and constantly gathers information from monitored www sources. Within seconds after information is available from the source it is analyzed, structured and available for further analysis as a transactional feed.

Does SENTRA monitor only company data?

No. SENTRA can monitor and extract transactions from any type of unstructured data. SENTRA “understands” many topics and industries to include: Economics, Finance, Banking, Engineering, Computer Science, Health Care, Pharmaceutical Industry, Telecommunication Industry, Science, Politics, etc. Please contact us and ask for more details.

How is SENTRA data supplied?

You can acquire SENTRA time-series transactions and reference data as Data as a Service (DaaS) via an API (XML, JSON or other format as required), via FTP or alternatively your SENTRA system can be entirely hosted (SaaS) and accessible via a number of query and analytical tools.

Sentiment Analysis has a reputation for not being too accurate, how does SENTRA perform?

To date, typical sentiment analysis engines are based primarily on statistical techniques. Most are not granular (article, not sentence level based), do not understand “context” or “relationships” and achieve accuracy rates below 70%. However, SENTRA’s scalable Sentic Computing based system routinely delivering accuracy rates in excess of 85-90%. This is equivalent to (even slightly in excess of) human performance.

How does SENTRA assure data quality of transactions?

There are two components to assuring data quality in sentiment based transactional output from heterogeneous, unstructured data sources such as news, blogs, social media, etc (collectively called – articles):

1. Entity Extraction has to be of high quality – to understand how Zoral Labs assures data quality of extracted entities from unstructured sources, please refer to Zoral Labs data quality management.
2. Sentiment-Entity extraction - Zoral Labs approach to management of extracted Sentiment-Entity data quality is as follows:

We assure sentiment quality across a number of specific domains such as: economics, business and finance, telecommunications and technology, healthcare and pharmaceuticals, food and beverages, travel, electronics, entertainment, retail, politics, etc. In addition to domain coverage, our data quality assurance testing is organized by different lexico-grammatical themes (conjunctions, simple sentences, conditionals, subjunctives, polarity unification rules for different grammatical structures, etc.).

Multi-domain Sampling

Our Knowledge Managers (KM) perform weekly random, multi-domain, statistically significant input/output sampling and review newly generated sentiment transactions for correctness. The reviewed articles are added to the regression test suite and SENTRA’s sentiment analysis system component is fortified to fix all incorrect transactions. Full regression suite run ensures that no prior tests are broken by the expanded sentiment analysis component logic and knowledge base. In addition to manual testing, KM marks a transaction as incorrect and the system is able to automatically find faulty system logic, fix it, run regression suite and report the auto-correction results – this method of data quality assurance is semi-automatic and is in addition to manual testing.

Regression Testing

The regression test suite consists of manually and semi-automatically analyzed articles for each domain and lexico-grammatical theme. Each time the sentiment analysis engine is enhanced, the test suite is used to ensure the change was virtuous. Sentra’s regression suite is constantly augmented by semi-automatic, multi-domain input/output sampling. Sentra’s English language regression test suite basic facts:

• 4 years of development
• ~12000 high quality, periodically reviewed, multi-dimensional test cases
• ~300 advanced tests are added every month

Can SENTRA be supplied with analytical tools?

Yes. SENTRA has a web interface so it can be easily accessed and integrated to a variety of search and BI analytical applications. Analysis results can be displayed in user defined Charts and Reports. Moreover, you can expand your monitoring capabilities by adding our advanced, vertical, scalable search engine to enable media search by keywords, tonality, time range, media types, Authors, Geography, Products, Companies, etc.

Who uses SENTRA?

Anyone who needs to monitor transaction changes to web data. For example, SENTRA is used by Finance Companies for monitoring client/market changes/risk, by PR Companies for monitoring client/product/competitor news, by Corporate Marketing Departments, Sales Executives and many others.

How much does SENTRA cost?

There are no capital outlays. You pay only for initial setup and the number of sites and items/companies you monitor. Costs depend on the number of sites and frequency of update required. Please do not hesitate to contact our sales office for a free quotation.

How quickly can I start?

Often, in as little as one week, depending on the sources and entities or fields you require. Typically you would give us details of the sources you require and the fields/event/relationship types to be monitored. We then provide a small, sample output for you to evaluate and then switch on your SENTRA system. If you require analysis of entities/relationships/events that we do not currently cover, then just provide us with samples and we can evaluate and incorporate them into SENTRA.

What next?

Simply contact our sales team by email or phone and we would be delighted to talk about your requirements.