Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/znerol/rejoha22-recommender
https://github.com/znerol/rejoha22-recommender
Last synced: 25 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/znerol/rejoha22-recommender
- Owner: znerol
- License: unlicense
- Created: 2022-11-25T16:21:02.000Z (almost 2 years ago)
- Default Branch: develop
- Last Pushed: 2022-11-26T10:22:22.000Z (almost 2 years ago)
- Last Synced: 2024-06-19T11:33:52.336Z (5 months ago)
- Language: Go
- Size: 298 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE.txt
Awesome Lists containing this project
README
REJOHA 2022 - Responsible News Recommender Systems
==================================================
This repository contains results for the
[Responsible News Recommender Systems](https://hack.opendata.ch/project/893)
challange at Rethink Journalism Hackaton 2022. Note that a responsible
recommender systems is an entire field of research in itself. This project
barely scratches the surface of it.
## Project goals
News sites often present a list of most-clicked content to their users. Since
some categories of content are more popular than others (e.g., News and Sports),
there is a risk that popular content from under-represented categories just
never makes it onto these lists.
This project explores ways to boost the diversity of categories represented in
most-clicked content lists.
## Data / Tools / License
Data is sourced from the SRG-SSR
[mostClicked](https://developer.srgssr.ch/apis/srgssr-video/docs/mostclicked)
endpoint for video content.
A developer account at SRG-SSR developer portal is necessary to run the software
and reproduce the results presented here. See the
[FAQ](https://developer.srgssr.ch/faq) for more info.
Source code is written in [go](https://go.dev/) and released under the
[unlicence](http://unlicense.org/).
## Step 1: Scrape data from SRG most-clicked video endpoint
Based on the aggregated popularity data (available for certain content on the
SRF APIs, for example), the first step would be to try to find out which
"categories" of content are over- or under-represented in the most popular
content.
### Top 200 videos
Generate a list of top 200 videos:
export SRG_OAUTH_TOKEN=xyz
go run cmd/top200/main.go > /tmp/top-200.tsv
Sample output:
Popularity Category Title Id
19483 Sport Doppelter Richarlison schiesst Brasilien mit Traumtor zum Sieg 516a880f-366f-43bb-9075-daccf1eddc1c
14244 News Tagesschau vom 24.11.2022: Hauptausgabe 60aa9e97-cfb1-48bb-951b-363107cb46a0
8059 News 10 vor 10 vom 24.11.2022 17c59931-8c51-4509-a3dd-6e67c3416b5f
7630 Sport Schweizerinnen kämpfen um EM-Gold fe0c0275-dbc6-4d3d-8441-9ebedf52df7c
6259 Kultur Mythos Matter 6702d57e-2f65-424d-8a0a-176a65b5e1f0
...
### Top 200 videos aggregated by category
Aggregate number of videos for every category:
export SRG_OAUTH_TOKEN=xyz
go run cmd/catpop200/main.go > /tmp/catpop-200.tsv
Sample output:
Popularity Category
66303 Sport
46909 News
28289 Unterhaltung
13639 Film
10798 Dok & Reportagen
10666 Kultur
9730 Wissen & Ratgeber
### Top 200 videos aggregated by views
Aggregate number of views for every category:
export SRG_OAUTH_TOKEN=xyz
go run cmd/catcount200/main.go > /tmp/catcount-200.tsv
Sample output:
Sum Category
38 Sport
31 News
20 Unterhaltung
19 Dok & Reportagen
18 Film
17 Serien
## Step 2: Boost under-represented content in most-clicked video list
The top hits of the most clicked videos are often from the same category
(sports, news). A simple way to boost content from less represented categories
to a better rank is to normalize the views within categories. Each entry is
given a score between 0 and 1, and the list is ordered by that score instead of
by the number of clicks.
### Top 200 videos with clicks normalized within each category
The most basic way to compute the score is to simply normalize the number of
clicks within the same category. The piece with most clicks in category A is
assigned a score of 1. The score of the remaining items in the same category is
computed by scaling down the number of clicks by the same factor.
This results in a list where each category is represented with exactly one item
at the top of the list (each of them with score 1).
Generate a list of top 200 videos (score by normalizing popularity within each category):
export SRG_OAUTH_TOKEN=xyz
go run cmd/topnorm200/main.go > /tmp/topnorm-200.tsv
Sample output:
Score Popularity Category Title Id
1.00 376 Kids Masha und der Bär - E churzi Reis 6b70748b-b89b-4bb9-9fec-aa6c63d291e6
1.00 1755 Talk Brian alias «Carlos» – Kommt er je frei? ec3704b5-cc9b-47cf-9781-6026c61d6471
1.00 323 Comedy Staffelfinale mit Kilian Ziegler (Staffel 14, Folge 10) 0f4565dd-5a7f-41b9-9034-566d1e89c60c
1.00 5608 Wissen & Ratgeber Wie wir Behinderte behindern 31f99f6e-d61c-46e1-b785-7d4c7ed9696b
1.00 3604 Unterhaltung Abschied mit Tränen e01df7e2-e401-4334-84c4-6a19ed807bd4
1.00 1147 Serien Vertrauen (Staffel 15, Folge 27) 1b981517-5a19-4fdc-9d68-6f010eef2ff1
1.00 1014 Gesellschaft Neuer TikTok-Star?: Sascha Ruefer geht mit WM-Kommentar viral 529bf8ab-7955-477d-bee3-5b9d9e752dd0
1.00 3402 Dok & Reportagen Wir, die Pflegefachkräfte von morgen (Staffel 1, Folge 1) 8bc3e03b-5088-4701-b860-f1726d9a0347
1.00 2984 Gelingt die Einigung im Rentenstreit? 9e8802ad-029e-4ce9-889a-ac3c9237a964
1.00 11112 Sport Auf dem Lift mit «Gumpiball» Aline Danioth bd1deb25-2ddc-43bd-b863-26a7c3b4e29a
1.00 11276 News Tagesschau vom 25.11.2022: Hauptausgabe 658b3c99-db7e-473b-bf17-717039cb98ac
1.00 2724 Politik Rentenstreit reloaded: Jetzt geht es um die zweite Säule 5f40dc56-253f-4a30-9872-996edf9b6088
1.00 1516 Kultur Black Friday: Kaufrausch vs. Konsumverzicht 86d70fa2-7951-4b34-a164-2bbff7d15edb
1.00 3195 Film Inga Lindström – Rasmus und Johanna 8af0ae11-102e-4a2d-8ffb-18e5aaed0ed4
0.90 10016 Sport Odermatt: «In Lake Louise gibt es nicht sehr viele Helfer» df9dcb12-461e-4a91-918e-38704eeaaaa1
0.85 977 Serien Hoffnungsschimmer (Staffel 15, Folge 28) b5b02651-d778-4596-98bf-1e69897b9517
0.80 1412 Talk mit Zarifa Ghafari 169e301e-d3bd-4e45-9b54-abcfe5d2049d
...
However, it is questionable if really every category needs to be represented
with one top-hit at the start of the ranking. Thus, let's try a mixed approach.
### Top 200 videos with clicks normalized using a mixed approach
A straightforward way to refine the score is to include more factors in the
equation. In addition to the score derived from normalizing clicks within a
category, a second factor is derived from normalizing clicks over all items.
To refine the contribution of one or the other factor to the total score, they
can be raised to the power of some exponent.
Generate a list of top 200 videos (score with mixed approach):
By default this is using the following exponents:
* Exponent used to weight the score derived from clicks normalized over all items:
`expGeneral=1`
* Exponent used to weight the score derived from clicks normalized within each category:
`expCategory=2`
Change the source code in order to adapt the exponents.
export SRG_OAUTH_TOKEN=xyz
go run cmd/topmixnorm200/main.go > /tmp/topmixnorm-200.tsv
Sample output:
Score Popularity Category Title Id
1.00 11276 News Tagesschau vom 25.11.2022: Hauptausgabe 658b3c99-db7e-473b-bf17-717039cb98ac
0.99 11112 Sport Auf dem Lift mit «Gumpiball» Aline Danioth bd1deb25-2ddc-43bd-b863-26a7c3b4e29a
0.72 10016 Sport Odermatt: «In Lake Louise gibt es nicht sehr viele Helfer» df9dcb12-461e-4a91-918e-38704eeaaaa1
0.50 5608 Wissen & Ratgeber Wie wir Behinderte behindern 31f99f6e-d61c-46e1-b785-7d4c7ed9696b
0.32 3604 Unterhaltung Abschied mit Tränen e01df7e2-e401-4334-84c4-6a19ed807bd4
0.30 3402 Dok & Reportagen Wir, die Pflegefachkräfte von morgen (Staffel 1, Folge 1) 8bc3e03b-5088-4701-b860-f1726d9a0347
0.28 3195 Film Inga Lindström – Rasmus und Johanna 8af0ae11-102e-4a2d-8ffb-18e5aaed0ed4
0.26 2984 Gelingt die Einigung im Rentenstreit? 9e8802ad-029e-4ce9-889a-ac3c9237a964
0.24 2724 Politik Rentenstreit reloaded: Jetzt geht es um die zweite Säule 5f40dc56-253f-4a30-9872-996edf9b6088
0.16 1755 Talk Brian alias «Carlos» – Kommt er je frei? ec3704b5-cc9b-47cf-9781-6026c61d6471
0.13 1516 Kultur Black Friday: Kaufrausch vs. Konsumverzicht 86d70fa2-7951-4b34-a164-2bbff7d15edb
0.11 2317 Film Inga Lindström – Hochzeit in Hardingsholm 123563f0-ec53-4c53-91f0-35cd19eee7ba
0.10 1147 Serien Vertrauen (Staffel 15, Folge 27) 1b981517-5a19-4fdc-9d68-6f010eef2ff1
0.10 5179 Sport Odermatt: «Hat mich selber etwas überrascht» fd74ec5b-612e-4fb8-84ef-f0eb9a1ff9f5
0.09 1014 Gesellschaft Neuer TikTok-Star?: Sascha Ruefer geht mit WM-Kommentar viral 529bf8ab-7955-477d-bee3-5b9d9e752dd0
0.08 1412 Talk mit Zarifa Ghafari 169e301e-d3bd-4e45-9b54-abcfe5d2049d
0.07 1191 Kultur Mythos Matter 6702d57e-2f65-424d-8a0a-176a65b5e1f0
0.06 977 Serien Hoffnungsschimmer (Staffel 15, Folge 28) b5b02651-d778-4596-98bf-1e69897b9517
0.06 1983 Dok & Reportagen Katar – Perlen im Sand c00cafb8-8a9b-47cb-9053-c294bd2815cb
0.06 4302 News 10 vor 10 vom 25.11.2022 8fb42def-a546-4c6a-b416-0d7d6e1bf959
0.05 1644 Politik Männer ausgeschlossen, Millionen für die Papst-Garde, Transfrau im Männergefängnis d20389af-8ee3-43a7-be9b-fd95ba74fb38
0.05 1943 Unterhaltung Darum sind die Lieder von Mani Matter zeitlos | Making-of | Neu aufgelegt: Mani Matter 2022 dfb3389b-52a9-4467-92a3-87e471f7e68e
0.05 1700 Oksana Mirinenko findet nur langsam zurück ins Leben 3851ea9f-83de-40e7-8096-459ded72f1e9
0.04 3735 Sport Feuz und Co. reut's, den Iltis freut's a745ab17-f042-4db1-b340-f7a717078124
### Effects of applying the mixed approach on the ranking
Evaluating the resultis it becomes clear that a higher value of `expCategory`
leads to more content of distinct categories among the top ranked entris.
Results for `expCategory=1`:
Results for `expCategory=2`:
