Liquid Helium is the core of all Stremor's products. A language processing system that uses heuristic models to extract information from text, Liquid Helium is the most feature rich language heuristics engine ever built. Liquid Helium extracts information about sentence and paragraph structure, word usage, parts of speech, grammar, writing style, punctuation, and author bias. This information provides insight into unstructured data and can be used in conjunction with Big Data system to unlock information, which was previously only minable through human data entry.

Liquid Helium offers all language analytics features through its API. It has the ability to summarize content to any length and at any level of detail. Extracting sentiment, bias, reading level, technical level, formality, urgency, and whether or not content is editorial in nature, are just a few examples of the language scores returned. Liquid Helium can also detect action items in the text. For example, if the author was looking for a follow up, the people mentioned, the time, the phone numbers, and the places.

Liquid Helium contains rules that range in the millions. Rules to determine if the writing style is formal or casual. Rules to determine if the author is for or against the subject matter. Rules to determine if the author is making an argument and if points made support the argument. It can even make inferences about the credibility of those points. This is possible through the vast amounts of data poured into its database about how words can typically be used and what they mean. 

Some may believe using heuristic science in language analysis infers it is a learning system. To some degree, it is. However, Liquid Helium is not sentient; it does not automatically learn on its own. To prevent accidental “learning” of bad habits, which can happen with learning machines, Liquid Helium has a modular design of interconnected pieces for normalization, extraction, classification, and sentiment that can be easily improved or modified for specific purposes. An improvement in one module enhances the performance of all others.

Stremor provides technology solutions that enable content platforms to support the evolution of online media towards an array of connected devices and systems. An intelligent content-aware foundation Liquid Helium addresses the needs of multiple screens in varying contexts and use-cases. As access to online content evolves to a broad ecosystem with thousands of combinations of screen sizes, access points, and use cases, content platforms need to be smarter.

Whether data mining or developing a knowledge base for internal use, organizations need to convert unstructured data from text, email, marketing materials, and support communications. The problem is, Big Data is turning the information from millions of new web pages, blogs, tweets, news articles, and comments, into actionable items for business decision making. However, Big Data only works for structured, normalized data. Creating structured, minable, and actionable data at web-scale is well beyond human capacity and requires automation. Liquid Helium delivers the core foundation of that automation.

Examples of the types of extraction to structured data and content tagging that can be performed by Liquid Helium include: 

Action Items: Extraction of action items from communications reader or external event handler. Finds appointment requests in emails and discovers completed tasks in support tickets.

People & Places: Identification of people and/or places mentioned in a document enables prioritization of emails, link generation, and categorization.

Keywords: Determination of keywords that define a document's subject matter.

Facts: Detection of key facts, statistics, quantities, and other verifiable data.

Reading Level: Determination of the reading level of a document based on vocabulary, sentence complexity, and other factors.

Date & Time: Extraction of time and date data from a document body or identification about relative temporal information (e.g. "last Tuesday" or "next Friday").

The Internet is a tremendous platform for free expression. It gives nearly everyone a means to express themselves. While social media can help organizations build communities, the reality of contemporary online culture are rich in superlatives, hyperbole, exaggeration, maliciousness, and outright lies. It is not enough to have a simple positive or negative sentiment analysis. An accurate gauge of online content and conversations requires consideration of sentiment, audience, writing style, and forum.

more than just positive or negative

Liquid Helium goes beyond simple recognition of trigger words and analyzes the context and broader meaning of each statement made by the author for an accurate normalized sentiment score. Additionally, it is able to determine if a comment is made in the context of an article, an editorial, or a forum post.  This context greatly influences how a remark should be interpreted, and how much impact it has on the audience. Combined with information about the biases of the author sentiment can rate not only the immediate sentiment, but changes in sentiment, detecting upswings and downswings in response to events.

context is everything

Sentiment analysis requires a basis of contextual information specific to a content category or business vertical. For example, the word “tiny” may indicate neutral sentiment in many topics, but would be negative as a description of a hotel room. Or overuse of exemplary embellishments in an article about an iPhone, on a website focused on Apple products, identifies the author having a high level of bias. This is why the Liquid Helium sentiment analysis is not a “one size fits all” solution. Each deployment is tailored to the subject matter and industry of the end user for an accurate range of sentiment scores.