mobileclick

mobileclick provides baseline methods and scripts for the NTCIR-12 MobileClick-2 task: http://www.mobileclick.org/

Requirements

Minimum requirements:

• Python 2.7
• NumPy
• nltk (>= 3.1)
• BeautifulSoup

Requirements for Japanese runs:

• mecab-python

Installation

Install mobileclick via PyPI:

$ pip install mobileclick

You can also install mobileclick from the source code:

$ python setup.py install

NLTK corpus download (for English):

$ sh nltk_download.sh

Mecab and mecab-python installation (for Japanese):

$ sh mecab_install.sh

Installed scripts

Download

Download MobileClick training data (Please sign up at http://www.mobileclick.org/ ):

$ mobileclick_download_training_data
Please input the email and password for http://www.mobileclick.org
Email: <Your email address>
Password: <Your password>

Download MobileClick test data:

$ mobileclick_download_test_data
Please input the email and password for http://www.mobileclick.org
Email: <Your email address>
Password: <Your password>

Baseline Runs

iUnit Ranking Subtask

Replicate the random iUnit ranking baseline:

$ mobileclick_random_ranking_method \
--runname random_ranking_method \
--query data/MC2-training/en/1C2-E-queries.tsv \
--iunit data/MC2-training/en/1C2-E-iunits.tsv \
--indexdir data/MC2-training-documents/1C2-E.INDX \
--pagedir data/MC2-training-documents/1C2-E.HTML

Replicate the LM-based iUnit ranking baseline:

$ mobileclick_lang_model_ranking_method \
--runname lang_model_ranking_method \
--query data/MC2-training/en/1C2-E-queries.tsv \
--iunit data/MC2-training/en/1C2-E-iunits.tsv \
--indexdir data/MC2-training-documents/1C2-E.INDX \
--pagedir data/MC2-training-documents/1C2-E.HTML \
--language english

iUnit Summarization Subtask

Replicate the random iUnit summarization baseline:

$ mobileclick_random_summarization_method \
--runname en_random_summarization_method \
--query data/MC2-test/en/MC2-E-queries.tsv \
--iunit data/MC2-test/en/MC2-E-iunits.tsv \
--intent data/MC2-test/en/MC2-E-intents.tsv \
--indexdir data/MC2-test-documents/MC2-E.INDX \
--pagedir data/MC2-test-documents/MC2-E.HTML

Replicate the LM-based iUnit summarization baseline:

$ mobileclick_lang_model_summarization_method \
--runname en_lang_model_summarization_method \
--query data/MC2-test/en/MC2-E-queries.tsv \
--iunit data/MC2-test/en/MC2-E-iunits.tsv \
--intent data/MC2-test/en/MC2-E-intents.tsv \
--indexdir data/MC2-test-documents/MC2-E.INDX \
--pagedir data/MC2-test-documents/MC2-E.HTML \
--language english

Replicate the LM-based two-layer iUnit summarization baseline:

$ mobileclick_lang_model_two_layer_summarization_method \
--runname en_lang_model_two_layer_summarization_method \
--query data/MC2-test/en/MC2-E-queries.tsv \
--iunit data/MC2-test/en/MC2-E-iunits.tsv \
--intent data/MC2-test/en/MC2-E-intents.tsv \
--indexdir data/MC2-test-documents/MC2-E.INDX \
--pagedir data/MC2-test-documents/MC2-E.HTML \
--language english

Generate your runs

Both of the subtasks are supported. Please look at examples in examples. A typical workflow is shown below:

1. Create a subclass of BaseRankingMethod/BaseSummarizationMethod

For iUnit Ranking

from mobileclick.methods import BaseRankingMethod

class YourRankingMethod(BaseRankingMethod):
    def init(self, tasks):
        '''
        Initialization
        '''
    def rank(self, task):
    '''
        Output ranked pairs of an iUnits and a score
    '''

For iUnit Summarization

from mobileclick.methods import BaseSummarizationMethod
from mobileclick import Summary

class YourSummarizationMethod(BaseSummarizationMethod):
    def init(self, tasks):
        '''
        Initialization
        '''
    def summarize(self, task):
        '''
        Output an instance of Summary.
        '''

2. Generate a run

For iUnit Ranking

from mobileclick import Task
tasks = Task.read(
    "data/MC2-training/en/1C2-E-queries.tsv",
    "data/MC2-training/en/1C2-E-iunits.tsv",
    "data/MC2-training-documents/1C2-E.INDX",
    "data/MC2-training-documents/1C2-E.HTML")
method = YourRankingMethod()
run = method.generate_run("YourRun", "This is your run", tasks)
run.save('./')

For iUnit Summarization

from mobileclick import Task
tasks = Task.read(
    "data/MC2-test/en/MC2-E-queries.tsv",
    "data/MC2-test/en/MC2-E-iunits.tsv",
    "data/MC2-test-documents/MC2-E.INDX",
    "data/MC2-test-documents/MC2-E.HTML",
    "data/MC2-test/en/MC2-E-intents.tsv")
method = YourSummarizationMethod()
run = method.generate_run("YourRun", "This is your run", tasks)
run.save('./')

Upload ./YourRun.tsv or ./YourRun.xml to http://www.mobileclick.org/ for evaluation!

License

MIT License (see LICENSE file).