A repository for some investigation and analysis into the long running, BBC Radio 4 topical discussion program: Any Questions.
First broadcast in October 1948, it was inteded to run for six editions only, this repository presents research and analysis into over 660 episodes (all of those available on the BBC Website).
To answer this question, from the data we need to extract where each Any Questions episode is coming from. Using Spacy, the NLP package for Python, I tried to extract locations from the text. I wrote some code to do this, then audited the first 20 episodes to check it's effectiveness. Below is a table of the problems encountered:
| Index | Problems |
|---|---|
| 2 | MISC swallows LOC inside of it |
| 7 | LOC is detected as PER |
| 11 | LOC is not detected |
| 12 | LOC is not detected |
| 17 | No LOC detected |
| 18 | MISC swallows LOC inside of it |
| 19 | LOC is not detected |
| 20 | LOC is detected as PER |
As can be seen from the table above, of the first 20 records I audited my code's functioning on; 12 (or 60%) had perfect location information) and the other 8's information wasn't usable. Further complicating this matter is that there is no method of determining what the location should be, no ground truth, which makes this either a very manual task or a technically infeasible one.
The wordlist: en_core_web_md might have better detection of place, potentially we should combine these for our purposes.
To answer this question, we need to use Spacy, the NLP package for python, to detect person entities in the body text. From an brief scan of the body texts for a couple episodes there seem to be one host and 4 guests listed, using this heuristic we might be able to able to improve our accuracy.
I wrote some code to extract names, and then audited the first 20 records to test it's effectiveness. Below is a table of the problems encountered:
| Index | Problems | Correct Names |
|---|---|---|
| 0 | PER didn't include Lord, Thought Brexit was a PER | 4 |
| 1 | PER as ORG | 4 |
| 3 | Sinn Féin as PER | 6 |
| 5 | ORG swallows PER inside of it | 4 |
| 6 | Wrong PER, LOCs in PER | 2 |
| 7 | LOC as PER | 5 |
| 8 | ORG swallows PER inside of it | 4 |
| 9 | ORG swallows PER inside of it, LOC as PER | 4 |
| 10 | PER didn't include Lord | 5 |
| 11 | ORG swallows PER inside of it, PER too short | 3 |
| 12 | ORG swallows PER inside of it | 4 |
| 15 | Sinn Féin as PER, PER not detected | 4 |
| 16 | ORG swallows PER inside of it | 4 |
| 17 | ORG swallows PER inside of it, PER not detected | 3 |
| 19 | LOC as PER | 5 |
As can be seen from the table above, of the first 20 records I auited my code's functioning on; 15 (or 75%) had problems to some extent and the other 5 were flawless. This clearly highlights just how difficult the extraction of names from the text is.
However the method detailed in the above section, is very good at extracting who hosted the episode - this happened flawlessly for all 20 records, so with this success it makes sense to apply it to the whole dataset and start to draw some conclusions.
| Host | Number of Episodes |
|---|---|
| Jonathan Dimbleby | 426 |
| Chris Mason | 124 |
| Shaun Ley | 27 |
| Ritula Shah | 27 |
| Eddie Mair | 20 |
| Edward Stourton | 8 |
| Ed Stourton | 8 |
| Nick Robinson | 6 |
| Martha Kearney | 5 |
| Anita Anand | 4 |
| Julian Worricker | 2 |
| Glenn Campbell | 1 |
Jonathan Dimbleby hosted the program from 1987 to 2019, taking over from John Timpson, in March 2019 he announced that he would "stand down... at the end of June." Shaun Ley, Ritula Shah, Ed Stourton and Julian Worricker served as interim moderators. In October 2019, Chris Mason was announced as Dimbleby's successor.
Below you can see I timeline I have created of who was hosting Any Questions:
Below is a bar chart displaying the days on which Any Questions was released:
We see that Friday is by far the most frequent release day, with Saturday being used only once!
Below is a bar chart representing who has produced each respective Any Questions episodes, this information isn't always available, so information should not be considered in an absolute manner but in relation to the other data points.
The first piece of information we have available about the Production team is the Producer, out of the 658 episodes we have details for the producer was credited just 371 times - just over half! Below we display a bar chart of those who have been credited as Producer:
We can see how some individuals have done a significant proportion of all Producing, with 20 different individuals credited so far, some like Lisa Jenkinson, Victoria Wakely or Camellia Sinclair have done the job sigificantly more than others. Let's next create a timeline to visualise who has been credited as the producer and when:
The timeline shows us a couple things, one of these is the bizarre pattern to which producers are credited. Victoria Wakely is consistently credited in the early 2010s, before we return to having no producer credited. Then Lisa Jenkinson begins to domain the Producing credits, however in recent years Camellia Sinclair has been credited ever so frequently. A final thing to note is that one Kathryn Takatsuk and a Kathryn Takatsuki appear in the list of Producers who have been credited, the former has done the job once whilst the latter has done so 3 times. I'm tempted to believe that these are the same person, however it's not like the BBC to make a mistake is it!
The next piece of information we have available about the Production team is the Lead broadcast engineer, out of the 658 episodes we have details for the Lead broadcast engineer is credited just 49 times. Below we display the chart of this:
We can see how many different individuals have been the Lead broadcast engineer, with 22 different individuals credited so far, some like Phil Booth have done the job significantly more than others.
The timeline shows us a couple things, the first is how infrequently Lead broadcast engineer is credited and the second is how they've only recently started being credited.
downloader.py- Download an episode, might be used laterepisode_details.py- Get details to all episodes and save it as a .pkl fileanalysis.ipynb- File for all the analysis we need to perform, using our data collected
From the documentation for the Stanford Named Entity Recogniser (NER):
- MISC, corresponds to miscellaneous entities, e.g., events, nationalities, products or works of art
- LOC, corresponds to location names
- PER, corresponds to person names
- ORG, corresponds to organisations