Question-Answering for Agricultural Open Data - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 23

(4) The user feedback that indicates which property was actually viewed is sent
to the server, and the corresponding mapping of the new

verb to the property

is registered in the KVS.
(5) Since the registered mappings are not necessarily correct, we recalculate the
confidence value of the mapping based on the number of pieces of feedback,
and update the ranking of the mapping to improve the N-best accuracy (see
section 4.3 ).

Also, we provide a registration mechanism of the user context information
to increase contents of the Plant Cultivation LOD. When the user registers
a sentence in the DB, the sensor data are automatically aggregated by using
built-in sensors on the smartphone, and the context information at that time
and location is inserted into the DB. For example, when a user registers a triple
describing a flower has blossomed, the sensor data for the location is converted
to literal, one for the temperature is converted to integer, and one for the space
is translated to Indoor or Outdoor, respectively. Then, the context information
such as gtcprop:flowerAddress (location), gtcprop:flowerDateHighTemp (high-
est temperature of the day), gtcprop:flowerDateLowTemp (lowest temperature
of the day), and gtcprop:flowerSpace (space of the flower) is automatically regis-
tered in the LOD DB. We prepared the LOD schemas (properties) correspond-
ing to the context information (see 'work logging' part of the previous listing).
Therefore, the user can register not only the direct assertion, but also several
background information at once.

4.2 Development of Question-Answering System

Figure 3 shows an interface of our question-answering system. It automatically
classifies the speech intention (Question Type) of the user into the following four
types (Answer Type is a literal, URI, or image).

1. Information Search
Search for plant information in the LOD DB.
2. Information Registration
Register new information for a plant that does not currently exist in the
LOD DB or add information to an existing plant.
3. Record Registration
Register and share records of the daily work. However, the verbs that can
be registered are limited to the predefined Properties in the LOD.
4. Record Search
Search through records to review previous work and view the work of other
people.

Figure 4 shows the architecture of our question-answering system. The user
can input a query sentence by Google voice recognition or keyboard. The sys-
tem then accesses the Yahoo! API for Japanese morphological analysis, extracts
a triple using the built-in dependency parser, and generates a SPARQL query by
filling in slots in a query template. The search results are received in XML format.