On

Conferences are an important part of scientific life and, even if I'm pretty new to this world, I understand planning one's tour at a conference can be a real hassle.

The context that led to the following project is pretty common for most scientists. My friend Fred Blain from USFD was preparing for the ACL2016 conference in Berlin. This year's ACL was a huge (1.6k researchers) conference with a lot of talks in up to 7 tracks in parallel. The program was announced on the website but, as usual, not easy to mine and dissect.

On a proposal from Fred, we decided to tackle the challenge of making this program easier to prune down to a small, personal interest list.

The rest of the article describes the parsing and restitution processes. The result can be found here.

## Parsing the HTML page

We'll go through the parsing process and the creation of a webapp to browse the program and generate an ICAL.

On the ACL website, some pages provide a detailed program (such as http://acl2016.org/index.php?article_id=64 ), we'll use that to generate a JSON list of events for each day.

### Setting up the basics

Let's first implement a base in which we could implement the parsing process:

from requests import get
from bs4 import BeautifulSoup

class ConferenceDay:

def __init__(self, day_id):
self.day_id = day_id
self.base_url = 'http://acl2016.org/'
self.session_soup = BeautifulSoup(get(self.base_url+'index.php?article_id='+str(self.day_id)).content, 'html.parser')
self.session_rooms_mapper = {
'A': "Audimax",
'B': "Kinosaal",
'C': "2002",
'D': "3038",
'E': "2094",
'F': "2091",
'G': "2097"
}


The class will hold everything, the room mapper is linked to the distribution of rooms at the venue.

### Page structure

First thing to do, as usual, is to study the page structure. One can easily show that the full page is actually a long table. Let's walk through the table's rows and understand the different features:

• The global events are in bold (tagged with <strong>)
• The sessions are preceeded by a <h3> title with the session's name
• Some useless events are present (Lunch & Coffee breaks)
• Some rows are here for padding only

Let's write a loop over the table's rows to classify them. As an output, we want a list of lists containing the row's content (as BeautifulSoup object) if it's a paper, the session name as a string if it's a title or a tuple (time,name) if it's one of the global activities.

def parse_html(self):
# list of events not to log
KILL_LIST = ['Coffee break', 'Lunch break']

# empty list of events and boolean to know if we're
# in one of the sets of activities (Lunch, Coffee, etc...)
self.list_events = []
in_activity_list = False

row = self.session_soup.find('tr')
while row:
# if it's a title, add a new sublist
if row.find('h3'):
self.list_events.append([row.find('h3').get_text()])
in_activity_list = False # beginning of session => end of global activity time

# if the second column is of length 1 => activity or empty row
# in beautiful soup the contents' length of a tag with 1 children only
# is of length 1
# the "empty" rows are 1 character rows => len()=1
elif len(row.find_all('td')[1].contents)==1:
# if strong => activity
if row.find('strong') and not row.find('strong').get_text() in KILL_LIST:
if not in_activity_list:
self.list_events.append([])
in_activity_list = True
self.list_events[-1].append(
(row.find('td').get_text(),
row.find('strong').get_text())
)

else:
pass

else: # it's a normal row => add it to the current list
self.list_events[-1].append(row)

# neeext !
row = row.find_next_sibling()


Of course, the else: pass is optional but is kept because readability counts. The function is straight forward, it just checks line by line for a title, a global activity or a normal paper and create sublist for sets of activities or sessions.

### From list to dict

Now we have the list, let's transform it into a dict mapping each timeframe with corresponding events.

def generate_dict(self):
self.events= {}
for sub in self.list_events:
# very long lists are posters or demo
if len(sub)>15:
continue

# check if the sublist is full string => global events
if len(list(filter(lambda _: type(_)==tuple, sub)))==len(sub):
for e in sub:
self.events[e[0]] = {'type': 'global_session', 'title': e[1]}

else: # normal session
# the first item was always the session name
session_title = sub[0]
# retrieve session room (last character of the session number)
session_room = self.session_rooms_mapper[session_title.split(':')[0][-1]]

for talk in sub[1:]:
local_timeframe = talk.find_all('td')[0].text
paper_title = talk.find_all('td')[1].contents[0].text
paper_authors = talk.find_all('td')[1].contents[2]
if not self.events.get(local_timeframe):
self.events[local_timeframe] = []
self.events[local_timeframe].append(
{'type': 'session',
'session_name': session_title,
'title': paper_title,
'authors': paper_authors,
'room': session_room}
)


This second function goes through the previously created list and build the dict:

• It first kicks out every sublist with more thant 15 entries (that corresponds to poster sessions or demo in the present conference).
• It creates a global event entry for each sublist composed only of tuple
• It generate a JSON entry with the right session name and room for other papers

Finally, don't forget to call those last two functions in __init__ if you want the list to be created at instantiation:

parse_html()
generate_dict()


In the end, the attribute events holds the full list of events, timeframe by timeframe.

## Bundling into a website

In order to make all this accessible and easily profitable, let's build a small webapp around it.

### Design considerations

The number of expected researchers was high for a conference but not for a webapp, we decided a simple plain JSON file could serve as a simple database. Note that this is not true in general and is highly correlated with my laziness.

We are bad at creating beautiful designs but HTML5 Up is not, we downloaded the style-sheet and HTML boilerplate from them (it's called Prologue). As the theming part is just a matter of integration it will not be discussed here.

### Generating the database

As we didn't want to kill ACL website by generating 3 requests to their servers every time 1 is sent to ours, let's cache the cleaned list of events and talks.

This is done simply using the following script (the comments should make it sufficiently readable):

from ACLConf import ConferenceDay
from json import dump

# filename where to store the final lists
DB_NAME = "./days.json"

# each day is associated to the corresponding article_id from ACL Website
mapper = {
'monday': 64,
'tuesday': 65,
'wednesday': 66
}

# create and fill a dict with day_name ->  event list by timeframe
days = {}
for day_name,day_id in mapper.items():
days[day_name] = ConferenceDay(day_id).events

# write it all in the "DB" file
with open(DB_NAME, 'w') as fh:
dump(days, fh)


### The webapp

As often in my projects, the webapp is built on top of Bottle. It consist in 2 functions:

• the first view, responding on / shows 1 form per day and allows one to pick up interesting talks and click on a Generate ICAL button to get the file (1 per day);
• the second one process the POST data from the form and generate the ICAL itself

from bottle import\
route,\
run,\
jinja2_template as template,\
request,\
response,\
static_file

@route('/assets/<filename:path>')
def static(filename):
"""Serves static files"""
return static_file(filename, root='assets/')



The second view to add is the index page which displays the forms. It's also the one that reads the JSON file:

from json import load

DB_NAME = "./days.json"

@route('/')
def index():

# keep a list of days so we can iterate in the right order
days_list = ['monday', 'tuesday', 'wednesday']

with open(DB_NAME, 'r') as fh:

# orders the sessions by ascending starting time
sessions_orders = {}
for d in days_list:
l = []
for time in days_contents[d].keys():
if len(time.split(':')[0])==1:
l.append(('0'+time,time))
else:
l.append((time,time))

# Good ol' sort (WARNING .sort() operates *in place*)
l.sort()
sessions_orders[d] = l

return template('index.html',
days_contents=days_contents,
sessions_orders=sessions_orders,
days_list=days_list
)


Some may notice a better way to sort session would be to use the built-in datetime module but here's the thing: we finished this webapp at 2.30 a.m. and we didn't want to mess around with this date-related crap.

The relevant part of the index.html template looks like this:

{% for d in days_list %}
<section id="{{d}}">
<div class="container">
<h2>{{d|capitalize}} sessions</h2>
{% for sorted_key,timeframe in sessions_orders[d] %}
<form action="/export.ics" method='POST'>
<input type="hidden" name='day' value="{{d}}"/>
<h3>{{timeframe}}</h3>
<ul>
{% for paper in days_contents[d][timeframe] %}
<li>
id="{{paper.title}}"
value="{{paper.title}}
{% if paper.type=='session' %}
by {{paper.authors}}#{{paper.room}}
{% endif %}
"
/>
<label for="{{paper.title}}">
<strong>{{paper.title}}</strong>
{% if paper.type=='session' %}
<em> by {{paper.authors}}</em>
{% endif %}
</label>
</li>
{% endfor %}
</ul>
{% endfor %}
<br />
<input type="submit" value="Generate iCal!"/>
</form>
</div>
</section>
{% endfor %}


It's literally just a couple of for-loops that spans one form per day and that add a new set of radio buttons for each timeframe.

### Generating the ICal

Because it was 2.30 a.m. and the first time for us with the icalendar package, we went for the easy way: reuse an example from the documentation.

In the following view, we get the POST data from the form and we process it. As all the required data was already included in the form itself (title, authors, rooms and timeframe) there's no need to call the DB again.

from icalendar import Calendar, Event
import pytz
from datetime import datetime

# use the right extension in the route to help the
# OS to detect it as a calendar file
@route('/export.ics', method='POST')
def generate_ical():

# init the Calendar object as proposed in the example
cal = Calendar()

# pre-init datetime objects for date handling
mapper = {
'monday': datetime(2016, 8, 8, tzinfo=pytz.timezone('Europe/Paris')),
'tuesday': datetime(2016, 8, 9, tzinfo=pytz.timezone('Europe/Paris')),
'wednesday': datetime(2016, 8, 10, tzinfo=pytz.timezone('Europe/Paris'))
}

for k,v in request.POST.items():
# if the field is named day it's the first (hidden) one
# it will be used afterwards but does not require a special action
if k=='day':
continue
else:
# update the pre-built datetime objects
start,end = k.split("-")
dtstart = mapper[request.POST['day']].replace(hour=int(start.split(":")[0]),
minute=int(start.split(":")[1]))
dtend = mapper[request.POST['day']].replace(hour=int(end.split(":")[0]),
minute=int(end.split(":")[1]))

# bundle everything into an event object
event = Event()
if v.find('#')>0:
# papers "value" in the form contain a # between authors and room
v_sum, v_room = v.split('#')
else:

# add it to the calendar and go to the next one

response.content_type = 'text/calendar'
# raw response without rendering
return cal.to_ical()


Aaaaannnnd... There we are!

A fully functionning ICal generator in less than 200 lines of code, accessible through a web browser.

## What's next?

As usual, the result we got is not an end, there's plenty of other conferences to come and it could be awesome to provide such a tool for every single one. This will require a better codebase, with a more modular architecture and a more efficient engine.

For those who think it's kind of overkill to use such a solution, some commercial alternatives exist and particularly Conference4Me. The tool proposed here doesn't serve the same purpose though: no application, no "branded" content nor support; just an idea, a need and a bunch of lines to fill it.