IEEE Computational Intelligence Magazine - November 2019 - 81

Communication Noise
Complaining
Make Noise

Commerce Buy
Commerce Scenario*
Getting*

Competition
Participation

Intentionally Act

The irony of Chelsea fans moaning about China buying all those players and ruining football.
Distinctiveness
Compatibility

People
Taking Sides

Preventing
Thwarting

Political Locales

FIGURE 3 Sentence (1) annotated with mappings from TransX. The green boxed frames represent the new annotations obtained by TransX mappings. The symbol (*) in front of Frame labels represents new profile specific frames on existing annotations.

adjective and verb synsets. Figure 2
shows the running example annotated
using DirectX.
3.3. TransX

It further extends DirectX using (i) transitive hyponymy relations, (ii) unmapped
siblings of mapped noun/verb synsets
and finally, (iii) derivational links. Figure 3 shows the running example annotated using TransX.
The above examples are annotated
using Word Frame Disambiguation (WFD)
framework which is a detour based
approach to frame detection. It can be
accessed through an API available online9.
It is implemented as a pipeline including tokenization, POS tagging, lemmatization, word sense disambiguation (using
UKB [39] and Babelfy [9]), and finally
frame detection by detour using the
WFD profiles described above.
4. Leveraging Frame Semantics for
Extracting Semantic Features

To extract certain semantic features of
each tweet, Word Frame Disambiguation
(WFD) API (see Section 3 for details)
was used for improving the results.WFDAPI takes text as an input and generates
an output with the corresponding BabelNet synsets and FrameNet frames along
with the position of the lexical unit evoking the frame as an output. In our experiments we did not apply any normalization
to the tweets, i.e. we did not try to transform hashtags into lemmas and we did
not correct the spelling of the tweets.
Even working on the raw tweets, we
were able to cover 96% of tweets with
9

http://lipn.univ-paris13.fr/framester/en/wfd_html/

extended semantic information. Semantic
frames and BabelNet synsets have been
extracted using the Base and TransX profiles as detailed in Section 3. Figure 4
shows the output of the WFD API for
the tweet Oh, you're 14 and quit smoking?
How inspiring!
We considered all the frames evoked by
each word in a tweet and created several
other new datasets from the combinations
of the following features: tweet, BabelNet
synsets and semantic frames. We augmented the original tweets with the semantic
features using the following strategies:
1) only tweets (our baseline)
Oh, you're 14 and quit smoking?
How inspiring!
2) frame annotations only
Process_stop Activity_stop Capacity
Ingest_substance Ingestion Breathing
3) BabelNet synsets only (using Babelfy
for Word Sense Disambiguation)
(replacement),
s13696060v s00084710v
s00093939v s00084027v

Process_Stop

bn:1369030v

4) both frame annotations and BN
synsets (replacement),
Process_stop Activity_stop Capacity
Ingest_substance Ingestion Breathing
s13696060v s00084710v
s00093939v s00084027v
5) tweets along with their frame annotations (augmentation)
Oh, you're 14 and quit smoking? How
inspiring! Process_stop Activity_stop Capacity
Ingest_substance Ingestion Breathing
6) tweets with their BN sysnets (augmentation)
Oh, you're 14 and quit smoking?
How inspiring! s13696060v
s00084710v s00093939v
s00084027v
7) data set containing three layers i.e.,
words, BN synsets and the frame
annotations (augmentation)
Oh, you're 14 and quit smoking? How
inspiring! Process_stop Activity_stop
Capacity Ingest_substance Ingestion
Breathing s13696060v s00084710v
s00093939v s00084027v

Activity_Stop

bn:00084710v

Oh, you 're 14 and quit smoking ? How

Breathing

bn:00084027v
inspiring!

bn:00093939v

Process_Stop

Ingestion

Ingest_Substance

FIGURE 4 Output of the Framester APIs with the tweet Oh, you're 14 and quit smoking? How
inspiring!

NOVEMBER 2019 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE

http://lipn.univ-paris13.fr/framester/en/wfd_html/

IEEE Computational Intelligence Magazine - November 2019

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2019