Depending of how do you get the data, I mean, if you get an unique raw string with all the messages from both speakers or you get the messages from each speaker separately.
A basic approach would be to establish the string "speaker X:" (where N is the speaker number) as the speaker tag for the first speaker then you could extract each message from each speaker using tools like NLTK and/or built-in functions like find().
Note: When I talk about a tag, I refer to some expression that would allow us to determine if a message is from a certain speaker or not.
Example:
You get the whole text that includes all the interventions of the speakers.
1) Set all speakers tags to distinguish their interventions in the whole text.
Example: The speaker tag for the first speaker could be "speaker 1:"
2) Find all the interventions of a speaker using str.find("speaker_tag")
3) Add all the interventions of each speaker to different data structures.
I think that a list of interventions of the speaker could be useful and then if you want to get all these interventions in one text message again, you could use
some built-in function like str.join() to joining them into one string again.
Other option to solve this would be using a tool like NLTK (I think this tool is great to classify text)
It has very useful features like tokenization that I think it's would be useful to solve your problem.
In the following example, I am going to use find() and slicing for a basic example about text tokenization:
Text data:
text = "speaker 1: hello everyone, I am Thomas speaker 2: Hello friends, I am John speaker 1: How are you? I am great being here speaker 2: It's the same for me"
Code example:
from itertools import islice, tee
FIRST_SPEAKER_TAG = "speaker 1:"
SECOND_SPEAKER_TAG = "speaker 2:"
def get_speaker_positions(text, speaker_tag):
total_interventions = text.count(speaker_tag)
positions = []
position = 0
for i in range(total_interventions):
positions.append(text.find(speaker_tag, position))
# we increase the position by the addition of all the previous
# positions to reach the following occurrences through the list of
# positions
position += sum(positions) + 1
return positions
def slices(iterable, n):
return zip(*(islice(it, i, None) for i, it in enumerate(tee(iterable, n))))
def get_text_interventions(text, speaker_tags):
# speakers' interventions of the text
interventions = { speaker_tag: "" for speaker_tag in speaker_tags }
# positions where start each intervention in the text
# (the last one is used to get the rest of the text, because it's the
# last intervention)
# (we need to sort the positions to get the interventions in the correct
# order)
speaker_positions = [
get_speaker_positions(text, speaker) for speaker in speaker_tags
]
all_positions = [
position for sublist in speaker_positions for position in sublist
]
all_positions.append(len(text))
all_positions.sort()
# generate the list of pairs that match a certain intervention
# the pairs are formed by the initial and the end position of the
# intervention
text_chunks = list(slices(all_positions, 2))
for chunk in text_chunks:
# we assign the intervention according to which
# list of speaker interventions the position exists
# when slicing we add the speaker tag's length to exclude
# the speaker tag from the own intervention
if chunk[0] in speaker_positions[0]:
intervention = text[chunk[0]+len(speaker_tags[0]):chunk[1]]
interventions[speaker_tags[0]] += intervention
elif chunk[0] in speaker_positions[1]:
intervention = text[chunk[0]+len(speaker_tags[1]):chunk[1]]
interventions[speaker_tags[1]] += intervention
return interventions
text_interventions = get_text_interventions(text, [ FIRST_SPEAKER_TAG, SECOND_SPEAKER_TAG ])
Notes:
If you have any doubt, you can read more details in the itertools documentation:
- Documentation about itertools.islice and itertools.tee:
islice
tee
Feel free to ask me anything you didn't understand about the example.
I hope you find it useful! =)
