When Big Data is not so big anymore

sasha uritsky

Oct 11, 2014

We are inundated with information. There is so much information around us they coined a special term - Big Data. To emphasize the sheer size of it.

It is, of course, a problem - to deal with a large amount of data. Various solutions have been created to address it efficiently.

At nmodes we developed a semantic technology that accurately filters relevant conversations. We applied it to social networks, particularly Twitter. Twitter is a poster child of Big Data. They have 500 million conversations every day. A staggering number. And yet, we found that for many topics, when they are narrowed down and accurately filtered, there are not that many relevant conversations after all.

No more than 5 people are looking for CRM solutions on an average day on Twitter. Even less - two per day on average - are asking for new web hosting providers explicitly, although many more are complaining about their existing providers (which might or might not suggest they are ready to switch or looking for a new option).

We often have businesses coming to us asking to find relevant conversations and expecting a large number of results. This is what Big Data is supposed to deliver, they assume. Such expectation is likely a product of our ‘keyword search dependency’. Indeed, when we run a keyword search on Twitter, or search engines, or anywhere we get a long list of results. The fact that most of them (up to 98% in many cases) are irrelevant is often lost in the visual illusion of having this long, seemingly endless, list in front of our eyes.

With the quality solutions that accurately deliver only relevant results we experience, for the first time, a situation when there are no longer big lists of random results. Only several relevant ones.

This is so much more efficient. It saves time, increases productivity, clarifies the picture, and makes Big Data manageable.

Time for businesses to embrace the new approach.

Interested in reading more? Check out our other blogs:

WHAT IS AI TRAINING

AI training is a critical part of conversational AI solutions, a part that makes AI software different from any kind of software previously created.
AI training is not coding.
Unlike all other existing software which is fully coded.

Let us consider a simple example:
We create chatbots for two companies, one company is selling shoes, another is selling cars. From the software standpoint it is one chatbot solution running as an online service accessed remotely or a program available locally. In both cases they are two identical instances of the same software (one instance for the shoes company, another for the cars company).
Yet, for the first company the chatbot is supposed to talk about flip-flops, summer shoes, high heels and so on. For the second company, however, the chatbot is not expected to know any of that. Instead, the chatbot should be able to support conversations about car brands, car models, should know how to tell Toyota Camry from Toyota Corolla, etc. This shoes and cars knowledge is not programmable. It is trainable. It is not coded, instead it is a part of language processing capability that AI solutions like chatbots have. And herein lies the major differentiation and advantage of the AI solutions compared to traditional software.

How to train AI?
There are several ways to do it. Sometimes AI system can train itself, improve its linguistic ability over time. It also can be trained by professional linguists. And in some cases, by the users. The latter is the desirable scenario because businesses know better than anybody else what they want their chatbot to talk about.
It is not easy, given the existing state of AI technology, and usually requires a high level of technical knowledge. You may have heard mentions of intents and entities in chatbot discussions. These are examples of linguistic elements AI training is currently based on.
Without proper understanding of what these linguistic elements are and how language acquisition process works in existing AI systems it is better to leave AI training to professional linguists.

Integrated Real-Time Data Boosts Content Delivery

How to make content more relevant and appealing to the content consumer?

This is a problem that has been on the mind of content creators for some time now. In our age of information abundance it is not easy to stand out and make your voice heard. The competition for the consumer’s attention is escalating, and with the number of information sources ever increasing, it will only get tougher.

Traditionally, a content delivery does not change across the target audience. A commercial, or a blog, looks and is experienced in the same way by all viewers and readers. We are entrenched in this paradigm, and can hardly imagine it being otherwise.

It turns out, the advancement of new technologies capable of capturing individual intents in real time brings up new opportunities in creating personalized experiences within the framework of content delivery.

This is how content can become more relevant - by becoming more personalized.

In a rudimentary form, we are already familiar with this approach as seen in online advertising. Some web and social resources aim at personalizing their promotional campaigns based on whatever drops of behavioural patterns and interests they can squeeze out of our web searches. The problem, of course, is that the technologies used to power these campaigns understand human behaviour poorly and results, therefore, more often than not leave a great deal to be desired. To put it mildly.

nmodes has been working on semantic processing of intent for several years. We now can capture intent from unstructured data (human conversations) with accuracy of 99%. (Interestingly, many businesses do not require this level of accuracy, being satisfied with 90%-92%, but we know how to deliver it anyway).

We recently started to experiment with personalizing content by using available consumer intent.

We used Twitter because of its real-time appeal.

We started by publishing a story, dividing it into several episodes:

And we kept the constant stream of data flowing, concentrating on intent to dine in Paris:

We then merged the content of the story with consumer intent to dine in Paris as captured by our semantic software. Like this:

This merging approach shows promising results - the engagement rate jumped above 90%.

Overall we are only at the beginning of a tremendous journey. We know that other companies are beginning to experiment, and the opportunities from introducing artificial intelligence related technologies into content delivery are plentiful.

There is a long road ahead, and we've made a one small step. But it is a step in a very exciting direction.