Geoscience Reference
In-Depth Information
inclusion and orientation. On the other hand, the process of extraction and indexing
of this system (see Table 2.1) only manages ASFs without taking into account the
possible associated spatial relationships. The difference of IR precision between
these two systems can be explained, in part, by this observation. It must also be noted
that the document collection used for the evaluation of the PIV prototype is smaller
in size and more localized (limited to the Pyrénées mountains) than that implemented
for the SPIRIT system evaluation. This gives an advantage to the results of the PIV
system in terms of precision. Moreover, let us note that during the evaluation of the
SPIRIT system, Purves et al. [PUR 07] show the contribution of a spatial IR in
complement to a thematic IR for answering these queries containing spatial and
thematic criteria.
The experiment relative to the temporal IR, presented in [PAL 10a], focuses on
theentirety ofthe 10books, inother words5,645 paragraphs (1,702TFs: 1,561 ATFs
and 141 RTFs). The protocol contains 35 queries: 15 queries focus on ATFs (five
with small temporal range such as “day”, five with intermediate range such as
“month” and five with large range such as “year”); 20 focus on RTFs (five for each
type: adjacency, inclusion, orientation and interval). Three people have conducted a
pooling-type assessment phase described in [PAL 10a]. The evaluation of these
results of PIV temporal IR gives a mean average precison (MAP) of 0.93. It was not
possible to compare this to other systems such as Mondeca (see Table 2.1), which, to
our knowledge, has not yet published test results of this type.
2.5. Summary
In this chapter, we have detailed our contributions concerning the spatial and
temporal information extraction, interpretation, indexing and retrieval from texts.
The main objective of these processes is the creation of indexes supporting spatial
and temporal IR scenarios in textual document corpora. The PIV prototype
implements these propositions.
2.5.1. Contributions
The PIV prototype supports four main modules dedicated, respectively, to the
extraction, interpretation, indexing and retrieval of spatial and temporal information.
The first two modules propose linguistic processes for the recognition, interpretation
and symbolic annotation of information. Two core models, respectively, spatial and
temporal, support this recognition, interpretation and representation, of toponymic
and calendar references but also of spatial and temporal relationships. The two other
modules are dedicated to the spatial and temporal information indexing and retrieval.
Thus, in this IR context, the symbolic annotations are transformed into numeric
representations organized into indexes essential to the spatial and temporal IR
models [GAI 08, LEP 07]. The PIV prototype shows the feasibility and usefulness of
such specialized IRSs.
Search WWH ::
Custom Search