Ich habe kürzlich eines für gemacht kabellose Ohrhörer. Viele von euch haben mich gebeten, eines für Kopfhörer zu machen, also hier ist es.
Kontext: Dies ist Teil meines Projekts, mit Reddit-Daten und LLMs herumzubasteln. Ich wollte etwas Nützliches für die Community schaffen und gleichzeitig meine Programmierfähigkeiten verbessern.
Die Idee besteht darin, hervorzuheben, welche Kopfhörer am meisten Anklang gefunden haben. Um es klar auszudrücken: Die meisten lieben =/= objektiv am besten. Aber hoffentlich ist es trotzdem ein nützlicher Datenpunkt, insbesondere für diejenigen, die von den Optionen überwältigt sind.
Offensichtlich ist dies eine sehr allgemeine Liste. Interessanter wird es, wenn Sie die Daten in Scheiben schneiden.
Ich habe zwei Folien, auf denen ich sie nach Rezensionen zu Musik vs. Spielen segmentiert habe. Wenn Sie tiefer in die Daten eintauchen möchten, können Sie dies unter tun Quelle / vollständige interaktive Liste
Sie können die Daten durchsuchen, die Kommentare lesen, nach Preis, Subreddits, kabelgebunden/kabellos filtern oder nach Kommentaren zu Musik, Spielen, Fitnessstudio, Laufen, Anrufen usw. filtern. Haftungsausschluss – die Seite enthält einige Affiliate-Links. Sie müssen sie nicht nutzen, sie helfen jedoch bei der Finanzierung der Analysen.
Methodik in den Kommentaren.
Von heyyyjoo
12 Kommentare
**Methodology:**
**Data collection:** I wrote a script that uses Google and Reddit search to search keywords like “best headphones”, “best headphones for music”, “best headphones for gaming” etc., filtered for the past year, sorted by relevance. I used LLMs (think ChatGPT but with preset instructions) to analyze each search result, extracting recommendations and reviews from the comments and performing sentiment analysis. I keep sampling search result page by search result page, until the cumulative relevant results analyzed dropped below 40%. A total of 936 relevant threads were analyzed, representing 7,601 users.
**Scoring:** Each user contributes up to 1 “vote” per model, regardless of how many times they mention it.
**Handling imprecise references:** If a user does not specify the exact model (e.g. “DT770 could refer to DT770 Pro, DT770 M, 80 ohm or 250 ohm version, etc.), their vote is still included. However, their 1 vote is “spread out” among the possible models. More popular/current models are given more weight (because the chance that those are the ones being referred to is higher).
**Ranking:** I calculated a final score that combines the net positive score (positive minus negative), and the positive-to-negative ratio (log applied to tame extreme ratio skew in models with few mentions). They are normalized and then summed up with a weight of 75:25.
**Improvements from previous versions:** Based on feedback about the importance of positive vs negative, I’ve changed the positive score component to net positive score. I’ve also added the percentage numbers for easier reference.
**Caveat:** handling and merging different model namings, abbreviations, and nicknames is non-trivial so a 100% LLM approach wasn’t sufficient. I did some eyeballing and manual clean-up but there may still be mistakes. Let me know if anything seems off or surprising!
**Tools:** Reddit API, Various LLM APIs, Firecrawl & Jina, Serper
[Source / Full interactive list](https://redditrecs.com/headphones/brand/beyerdynamic/)
I own the top one and agree with this sentiment.
Bose QC at 12? That’s too low
The ATH M50 has really fallen off from the decade where it was so frequently recommended that it was memed
the momentum 4’s sound great, but they have some of the worst software ever.
Just so you know the #2 in gaming shouldn’t be sony maxwell but audeze maxwell.
Guess no one likes skullcandy crusher evo? I prefer these over my soundcore
Would be better to have a Positive : Negative ratio.
Dividing P/N would give Sony a score of „3“, Sennheiser over „7“ and Soundcore „12“.
Why are you sorting by number of reviews and not percentage of positive reviews?
can you give a link to the earbuds analysis?
Surprised that CMF headphone pro is not there
Off-topic, but I just wanted to say thank you for these posts. My grandmother wanted to buy my dad a robot vacuum for Christmas and I know nothing about them. I used your post from about five months ago to help figure out which one to get. It was incredibly helpful! Thank you!