Information Technology Reference
In-Depth Information
the system. In particular, it provides the keywords that generate significant amount of
interest at the current point in time and the original text of the messages that
correspond to the keywords selected by the user. The Alchemy API is used for
affective analysis of the content of each Twitter message used by the system.
TinySong offers a simple interface for collecting links to the music tracks that match
with the keywords selected by the user. It is used for identifying the tracks that the
user wants to enter in his playlist and to collect additional info on these tracks that
will be used for retrieving them from YouTube. LastFM API is used for retrieving
similar music and artists to the ones selected by the user in her playlist while
YouTube is used to reproduce the tracks selected by the user. In addition, TwitterFM
uses its own rudimentary API for coordinating the use of the five APIs described
above thereby allowing the user to view/change Twitter trends and/or look for and
suggest appropriate music. The system offers appropriate endpoints for other
applications to use the results of TwitterFM in these areas. Finally the system uses the
FreeTTS API for converting the text of Twitter messages to speech. TwitterFM seeks
to recreate the radio experience when listening to Twitter messages. To this effect it
uses a radio metaphor for structuring user interaction. In particular, the system uses an
analog radio description of the keywords that are trending in Twitter and the ones
selected by the user (see Fig. 1). The messages received by the system are organized
in 'bands' similar to the way radio stations are organized in radio bands (FM. MW
etc). One of these bands called 'Trends” is generated automatically by the system
based on the top trending keywords in Twitter at each point in time. The rest of the
bands are determined by the user. In particular, the user selects the keywords that will
be included in each of these bands and these keywords appear on the radio screen as
'frequencies'. If the user selects any one of these keywords in the band then the system
moves the needle on top of the selection and the system starts collecting Twitter
messages containing this hashtag. These messages are then converted into speech
with FreeTTS and their affective content is analyzed using Alchemy. A tweet is
labeled as 'positive' or 'negative' and the label is accompanied by a numerical value
indicating the confidence of Alchemy for the label. Depending on the label and its
confidence TwitterFM colors each text either red for negative or green for positive
along with all the intermediate shades depending on the level confidence. For
example in Fig. 1 the middle tweet has been labeled as positive with confidence 0.8
(bright green) compared to the left one which was labeled as positive with confidence
0.2 (light green). The user can create a list of music tracks that wants to be played
along with the tweets. The systems plays sequentially the tracks in this list and also
allows the user to search for tracks and edit its playlist. TwitterFM constantly
monitors the number of remaining tracks in its play list and if this number is less than
a threshold it tries to append to the list tracks and artists similar to the ones in the
playlist using the LastFM API.
Users found the system original and interesting. The major problem they had
involved the actual content of Twitter messages which very often contains web-
specific information such as links to web sites that when spoken can be distracting. A
solution could involve screening the textual content of each message and removing
these links from the spoken part. In addition, due to the limitations of the TTS system,
TwitterFM ignores all punctuation marks in the message content. This results in a
spoken interpretation of the message that is flat and monotonous. Both problems
