Our latest release includes the following service enhancements:
- Reduced-cost named entity recognition: A solution targeted to applications that only require named entity recognition is now available. This uses the new recipe ner in the Sense Analysis service. It provides nearly the same accuracy as the defaultPrecision recipe but at half the cost. The named entities found are linked to the Language Graph and external references such as Twitter and Wikipedia. Learn more…
- Query rewriting controls: Several new controls were added to specify the type of transformations and filtering desired. Learn more….
Idilia is releasing today a semantic matching service. This new API determines whether one or more word senses are present in a document. For example given a Tweet with the word “apple”, is it related to the company or to something else?
How It Works
The application supplies a document and one or more target senses. A Sense Analysis is first performed on the document. The senses found are then compared to those provided using an advanced matching logic that considers sense equivalencies. Applications issue requests using our cloud-based API. The service capacity is fully scalable and adjusts to the demand.
This new service has numerous applications:
- Social Media Monitoring: Matching can eliminate most of the noise in Tweets or other social media. An application can identify an initial analysis set based on keywords and then refine the analysis set using the matching service. See our example for “detergent Tide” in Social Media Monitoring. Or try the online demo Twitter Filtering.
- Search Applications: Matching can be used to improve precision by removing false hits where the search keyword does not appear with the correct word sense in the candidate document.
- Profanity filters: Matching can identify if a document contains profanities by matching all the senses against a set of a few hundred profane word senses.
The Language Graph now includes external references to several thousand Twitter verified accounts. These new external references add to the existing external mappings to Wikipedia, IMDB and MusicBrainz.
The availability of external references against the Language Graph fine senses is a key enabler for applications embedding enriched content within documents or presenting relevant information on Internet search queries. Using the Sense Analysis API, developers can now map terms in a text to the corresponding Twitter accounts and display recent Twitter content.
Our recent service update includes a new capability to link to Twitter accounts (see External References to Twitter Verified Accounts. It also includes the following enhancements:
- Sense Analysis: The Semdoc format returned by the Sense Analysis service has been enhanced to include the Named Entity (NE) type and subtypes for proper nouns as part of the “SenseInfo” element. This simplifies applications and avoids querying the KB for this information. For more information, consult the Semdoc XSD.
- Paraphrase: We’ve made numerous improvements to the product search recipe. The quality of the paraphrases has been significantly improved.
- Language Graph: The API now supports requesting the Named Entity types and subtypes for common words connected under them in the Sense Property Hierarchy. For example, malaria/N1 is connected under NESUBTYPE_health_problem/N1. See the description for response element “neInfos” in Query Language. This can be useful to applications that attempt to classify the senses obtained using the Sense Analysis Service.
- Rate limiting: The Sense Analysis and Paraphrase services are now rate limited to prevent abuse. The default is 100 simultaneous requests. In batch mode, HTTP response header “Retry-After” is provided for the application to throttle itself while maximizing the throughput.