Search on anything... fitness, replay value ...
  • on android
  • on iPhone
  • on iPad
  • on iPad+iPhone
  • on windows phone
  • on all devices
GO
About us TipSense Platform In the News Contact
About the TipSense technology platform

Differentiating with Unstructured Data

TipSense is a next-generation technology platform designed to deeply capture meaning and insights from vast amounts of unstructured text content (reviews, blogs, tweets, messages, articles, etc...) across any domain area and convert it into actionable intelligence. Combining our breakthrough capabilities in statistical NLP, machine learning and patent-pending content fingerprinting technologies, TipSense can establish and maintain a dynamic graph of the complex relationships that reflect real world interactions. Our generative model can be used in a variety of applications, including enhancing recommendation and review systems, disambiguation, functional search, predictive analytics, affinity targeting and campaigns.

While significant advances have been made with big data analytics and structured data, the next generation of deep learning systems will also require the capability to harness the deep insights buried in unstructured data.

TipSense Applications

Discovery

In the creation of AppCrawlr, TipSense performed a deep analysis of content and signals from numerous app related sources and other tangentially associated domains. TipSense’s statistical engine was able to capture the full spectrum of how users actually engage with apps including:
 
  • Psychographics: (first time moms, amateur astronomers, frequent travelers)
  • Aspirational: (losing weight, passing time, save money)
  • Niche genres: (hack-n-slash, mid-core gaming, tilt shift filters)
  • Aspects: (replay value, engagement, ease-of-use, reliability)
  • Contextual: (for road trips, at doctors office, having a bad day)
  • Functional: (multiplayer, cloud sync, customizations)

To better understand how this works, imagine searching for a shooting game. TipSense's statistical analysis discovered a cluster of functionally related apps around the concept of 'hack-n-slash', based on aspects of 'combat', 'fast paced', and 'fighting games.' TipSense can then present this niche category of apps to users who otherwise might not be familiar with this genre. TipSense can further help a user navigate 'hack-n-slash' games, by exposing automatically inferred aspects, such as 'character customization' and 'upgrading weapons' and discover similar apps based on the related user segments of 'rpg fans' and 'casual gamers.'

These actionable insights power AppCrawlr’s guided discovery, which enables users to continuously refine their searches to discover the most relevant group of apps. For example, a search for calculator suggests specific types of calculators such as ‘scientific’, ‘unit conversion’ or ‘real estate’, specific features such as ‘big buttons’, as well as different target audiences such as ‘students’, ‘teachers’ or ‘investors’. The deep semantic understanding combined with unique digital fingerprints provided by TipSense allows AppCrawlr to make app recommendations based on countess functional, aspirational, genre, psychographic, and feature-based aspects.

Recommendation

Most leading recommendation systems are based on some variant of collaborative filtering using structured data (ratings, social signals, downloads, etc…). As an example, a movie recommendation would be largely based on behavioral and rating data of similar users for a particular genre (i.e. comedies)

Realizing that everyone has a unique taste profile, TipSense, is able to go deeper and capture all meaningful aspects to help explain why a user enjoyed a film. These latent aspects can range from ‘witty dialog’, ‘surprise ending’, and ‘extended car chase scenes’ to ‘cult following’, ‘detailed character backstories’ and 'cheer you up after a breakup.’ By leveraging the rich generative model from mining social and unstructured text, TipSense can begin to decipher and understand nuances of an individual’s implicit and explicit taste preferences, which allow for a higher degree of relevance and personalization in recommendation.

Targeting & Prediction

TipSense can enhance predictive modeling systems by exposing numerous insightful signals and features that can only be found from a deep semantic understanding of unstructured text. The deep concepts and relationships mined by TipSense can then be used to better target and improve ROI and performance of campaigns.

For example, a ‘restaurant deals’ campaign could improve relevance and conversion by having the ability to target at a more granular and meaningful level. TipSense uncovers predictive features that represent individuals' taste preferences all the way down to the dish and ingredient level and every detailed aspect of the establishment.

These factors could include aspects such as restaurants that specialize in 'savory meats' that are 'off the beaten path' with 'exceptional waiter service' and 'specialty vodka-based cocktails'. The capability to harness enrichment signals at this granular level can greatly enhance the predictive capability of targeting systems and generate lift while improving relevance to users and building trust and engagement.

Search & Inference

TipSense’s statistical knowledge extraction can effectively read between the lines to make inferences from captured concepts in a semantic way. By leveraging context from orthogonal domains, TipSense offers an alternative to keyword search by statistically inferring concepts, even when a term isn’t explicitly mentioned. For example, content fingerprinting can enrich a user query for ‘social apps’ with aspects of ‘making new friends’, ‘meeting new people’, and possible subtypes of ‘flirting’, ‘dating’, ‘relationships’, etc… which can power more meaningful query expansion and concept resolution for search. TipSense can extract additional meaning from user generated content and other content sources for functional search.

Similarly, content fingerprinting can be used for query disambiguation to help get to a user’s underlying intention. TipSense can, for example, differentiate a query for ‘jazz’ in the context of apps for music theory vs. ‘jazz’ for listening to music.

Search can also be enhanced with TipSense’s clustering technology to group results by functionally similar aspects, which can substantially improve diversity of results.



Unstructured text from any source


TipSense can be applied to numerous areas from social, travel, local deals, maps, coupons, books, movies, TV, music and news to software and enterprise content management systems. In each of these areas, TipSense can discover and represent the deep concepts, relationships, and nuances that can only be found in unstructured data. This extracted knowledge can be used to improve information retrieval and discovery, predictive analytics, social graph enrichment, functional search, and recommendation and targeting systems.
SEND FEEDBACK
Please tell us what:
  • You like
  • You didn’t like
  • Features you want improve
  • Ideas you might like to see
  • Wrong with this version
  • etc...
Email (optional):