Swedish Saplo Debuts "Semantic Search" at TechCrunch Disrupt

EricMack 0 383 Views

Swedish start-up Saplo wants to put the entire online world into context to help slice through the deluge of irrelevant information - and they're trying to do it by shifting from syntax-based search to algorithms based on high-profile academic research that understand the semantics underlying all that overwhelming amount of text on the Internet.

Saplo's latest effort to that end is Saplo Stream , which went live in an alpha phase Tuesday afternoon, right in the middle of in San Francisco. Saplo Stream's aim is to deliver only the most relevant news headlines and blog posts to users, weeding through all the slush and saving time.

We caught up with Saplo's Founder and CEO Mattias Tyrberg at TC Disrupt to get the scoop on the new service.

Daniweb - Really briefly, explain how Saplo Stream works - sell us on why someone would want to opt for it over some of the other many services that say they can filter the universe of online information for us.

Tyrberg - In this alpha release we are targeting the people using Twitter to read news. Say you have been away for one week and you have a meeting in 20 minutes, how can you possibly know what have happened with services such as Twitter and Google Reader? We solve this problem by adding a time and relevance filter on your news stream. Saplo Stream is about getting rid of information overload and letting you have the power of your news reading.

Daniweb - What's the name mean?

Tyrberg - Saplo stands for Semantic Application. Saplo Stream is for Semantic Stream, a personal stream based on semantics.

Daniweb - Your company is all about automated text analytics - can you tell us anything about your core analytic technologies? What sets it apart from Google or any of the other big players in that space? What's the difference between semantic analysis and what's out there now?

Tyrberg - Our technologies are based on research that mimic how the brain analyzes information. So, basically what it does is in some way understanding the meaning of a text and the relationships to other texts.

There are a lot of companies that do text analysis but not that many have an API. The main competitors are OpenCalais - but they only do tagging - and OpenAmplify, but they do not have the context analysis (text recommendation) capability.

Daniweb - What kind of user are you targeting? Is this just for total information junkies/journalists/high-volume information consumers? Or could your average person who just scans e-mail and headlines each day find it useful?

Tyrberg - Right now - Twitter users that want to be able to keep track of what's happening even if they don't read their Twitter stream every day. Since this is an alpha, the targeting is smaller but later on when we add more features we believe this can generate value for almost anyone.

Daniweb - What can you tell us about your roll-out strategy / funding, etc...

Tyrberg - Since this is our alpha release, we will continue to roll out new features during the year. Other than that we do have a text analysis API for companies that want to analyze their information today.

Earlier this year, Saplo completed a $500,000 seed round for further development of a text analysis platform from experienced industry leaders such as Professor Göran Grosskopf (the chairman of IKEA's holding company). One of the next steps would be increasing our presence in the USA.

saplo-at-techcrunch-disrupt-2010.jpg 42.4 KB

Dani AI

Generated

A few up-to-date notes to put the original Saplo Stream piece into practical context and help anyone trying to rebuild or replace that workflow today.

Saplo’s early work on semantic text analytics was later folded into a larger publisher/recommendation play after an acquisition by Sprinkle (now operating as Strossle) in 2015; the startup’s original alpha product from 2010 is not the current way to access that technology. (tech.eu)

“Semantic” here is best read as a move away from pure keyword matching toward concept-level matching: entity extraction, topic discovery and, crucially, vector representations of text that let you measure meaning-similarity. Marketing language like “mimic how the brain” (as quoted in the original interview) is shorthand; the practical building blocks are NLP pipelines plus embeddings / sentence encoders. (techtarget.com)

If the goal is a modern “semantic stream” (filter a social feed into a short, relevant digest), a compact, production-friendly pipeline is: ingest the feed via the platform API, normalize and chunk items, compute embeddings (OpenAI embeddings or SentenceTransformers models), store vectors in a vector DB for fast k-NN, rank by a relevance function that mixes recency with embedding-similarity, then dedupe/cluster and produce short summaries for presentation. Pay careful attention to API rate limits, privacy/retention rules and cost. (platform.openai.com)

If building this is too heavy, managed text-analytics and content-recommendation options exist: Amazon Comprehend, Google Cloud Natural Language and Azure AI Language provide entity/sentiment/summarization primitives, and publisher-focused vendors offer turnkey recommendation stacks (the acquirer integrated Saplo tech into that space). Choose based on data residency, latency, and whether you need an embeddings-first retrieval architecture. (docs.aws.amazon.com)

Credit: original thread by and the quoted comments from @tyrberg helped frame the problem; the notes above aim to translate that 2010 discussion into concrete, current engineering options.

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.

Edit Preview