For people suffering with obesity, logging calorie counts
and different dietary facts at each meal is a proven manner to shed pounds. The
technique does require consistency and accuracy, but, and when it fails, it's
generally because people don't have the time to find and report all the
information they want.
a few years in the past, a crew of nutritionists from Tufts
college who were experimenting with cell-telephone apps for recording caloric
consumption approached individuals of the Spoken Language structures group at
MIT's computer technology and artificial Intelligence Laboratory (CSAIL), with
the idea of a spoken-language application that might make meal logging even
less difficult.
This week, on the international convention on Acoustics,
Speech, and signal Processing in Shanghai, the MIT researchers are supplying an
internet-based prototype in their speech-managed nutrients-logging gadget.
With it, the user verbally describes the contents of a meal,
and the machine parses the outline and automatically retrieves the pertinent
dietary records from a web database maintained by means of the U.S. branch of
Agriculture (USDA).
The information is displayed collectively with pictures of
the corresponding foods and pull-down menus that permit the consumer to refine
their descriptions -- choosing, as an instance, precise portions of meals. but
those refinements also can be made verbally. A user who starts by announcing,
"For breakfast, I had a bowl of oatmeal, bananas, and a glass of orange
juice" can then make the amendment, "I had half of a banana,"
and the device will replace the facts it presentations about bananas at the
same time as leaving the relaxation unchanged.
"What [the Tufts nutritionists] have experienced is
that the apps that had been out there to help human beings try and log food
tended to be a bit tedious, and consequently humans failed to maintain up with
them," says James Glass, a senior research scientist at CSAIL, who leads
the Spoken Language structures institution. "so they had been seeking out
ways that have been correct and easy to enter statistics."
the first writer on the new paper is Mandy Korpusik, an MIT
graduate student in electric engineering and computer technology. She's joined
through Glass, who is her thesis advisor; her fellow graduate scholar Michael
charge; and with the aid of Calvin Huang, an undergraduate researcher in
Glass's group.
Context sensitivity
inside the paper, the researchers document the consequences
of experiments with a speech-recognition machine that they advanced specially
to deal with meals-associated terminology. but that wasn't the principle focus
of their work; certainly, an online demo of their meal-logging device rather
makes use of Google's loose speech-recognition app.
Their studies concentrated on two other problems. One is
figuring out words' useful position: The gadget desires to understand that if
the consumer statistics the phrase "bowl of oatmeal," dietary
statistics on oatmeal is pertinent, but if the word is "oatmeal
cookie," it's now not.
the alternative trouble is reconciling the person's
phraseology with the entries inside the USDA database. as an instance, the USDA
information on oatmeal is recorded below the heading "oats"; the word
"oatmeal" indicates up nowhere inside the access.
To cope with the first trouble, the researchers used gadget
mastering. through the Amazon Mechanical Turk crowdsourcing platform, they
recruited workers who really described what that they had eaten at current
meals, then categorized the pertinent words inside the description as names of
meals, quantities, logo names, or modifiers of the meals names. In "bowl
of oatmeal," "bowl" is a amount and "oatmeal" is a
food, however in "oatmeal cookie," oatmeal is a modifier.
once they had roughly 10,000 labeled meal descriptions, the
researchers used machine-gaining knowledge of algorithms to locate styles
within the syntactic relationships among words that might become aware of their
functional roles.
Semantic matching
To translate among customers' descriptions and the labels in
the USDA database, the researchers used an open-source database referred to as
Freebase, which has entries on greater than eight,000 not unusual food objects,
many of which encompass synonyms. wherein synonyms have been lacking, they
again recruited Mechanical Turk employees to deliver them.
The version of the gadget offered at the convention is
intended chiefly to illustrate the viability of its method to natural-language
processing; it reviews calorie counts however doesn't yet overall them
mechanically. A model that does is inside the works, however, and while it's
entire, the Tufts researchers plan to behavior a user study to determine
whether or not it indeed makes nutrients logging easier.
No comments:
Post a Comment