The set of incentives means the more successful you are, the more you will directly be attacked in the form of bad faith posts. Reddit is not better because it is a better system or because it has better moderation. Reddit is better because there isn't an entire industry (SEO) devoted to corrupting Reddit posts yet. Even google hasn't done a stellar job of winning that arms race, and they have vastly more resources.
You add value by performing the manual labor of finding and "reading" through posts, but the core value is generated by Reddit's abuse prevention team. If you were to be acquired by Reddit, my trust would go to zero. If you took outside investment, my trust would go to zero. My trust is already a little low because wirecutter specifically avoided items they couldn't get affiliate links for and now they are completely corrupted by new ownership. Wirecutter was the last review site that I used to make a purchase decision.
More than anything, I am worried that the dark mirror of this website, the "reddit optimization" industry is probably just as advanced and validated by efforts like this.
The only stable solution is for each person to curate and maintain their own set of sources, so that there are no high-value metrics for marketers to target. Exactly the opposite of what this service is trying to do.
I hate that it is in my interest for this to never become popular. It's a situation where my faith in the listed products is inversely proportional to the popularity of your site.
It's just an unfortunate alignment that goes against the core thesis of the (great!) idea.
weight ~= max(num_months_existencenum_commentscomment_score, 0 account not active in the past year) or somesuch, obviously adjusted.
Back when searching "best [product] 2014" was actually a way to find good stuff on the internet - fivestar was able to generate those answers on-the-fly. I'd find I'd spend an hour or so researching something like a router, plug "router" into fivestar, and my pick would be at the top of the list. Those days are gone for the same concerns others have raised for your site. fivestar worked until reviews started to get gamed.
This is an awesome website and I see myself using it. I hope it doesn't meet the same fate!
That's great that you are taking that into account, but I'm not totally sure how to interpret the value bars. So I presume a high value means lots of positive engagement. But how does a lot of negative engagement show up? I think it's valuable to know when there is a lot of chatter about something and it's mostly negative. Or when it's controversial.
Maybe something like a statistical distribution graph or weighted color gradient, so you could tell at a glance the density/quality/depth of discussion, and also the distribution of sentiment.
If there was a "salty old mfer posting on a forum in 2010" search I'd be much more inclined to use that. Google used to do ok for finding those but the SEO arms race killed it.
How far back are you searching? I'm noticing products on the top of a lot of these lists that were popular 10 years ago.
I am also curious what sort of spam filtering mechanisms you have in place? Just the spam filters before content ever hit sentiment analysis or relevancy analysis was 98% in my data. I imagine Reddit is better than Twitter, but there is still is going to be spam. What measures do you have in place and do you determine them?
Do you take upvotes into account with weighting reviews? That was/is a concern I had when working with reddit data. I used retweets as a proxy for sorting popularity, but not any other weighting.
I'm definitely interested in this segment as I've been doing it for a long time, if you want to talk please reach out my contact is in my profile.
P.S. Are you raising money from angels? Drop me a note if you are. [email protected]. I really need something like this!
Feedback for fine tuning your analysis: Cyberpunk 2077 is probably on top of PC gaming for the wrong reasons
So yeah, I like it.
The picture shows it does have armrests though. It would be great if you could have some button there to report incorrect information for each product.
https://looria.com/search?category.subcategory=Bug-Killing+G...
A feature request - I’d like to be able to adjust the time frame. Things like basketball shoes move a lot year to year, and it can be hard to find ones from 18 months ago.
Let me add some context:
Many people add "Reddit" to their search queries to find authentic product reviews. We fine-tuned a BERT model to extract product mentions from over 4 million Reddit comments and posts with Named Entity Recognition (NER). The result is a list of the most popular products across many subreddits.
No platform (including Reddit) is resistant to fake reviews and spam, but we think it's happening less frequently here for various reasons:
- Redditors and other forum members are more interested in boosting their ego by showing their depth of knowledge on the topic (and correcting others on the topic), whereas corporate websites are more interested in raking profit by displaying (potentially) dishonest information.
- Enthusiasts in subreddits are pretty good at spotting dishonest or fake content, which results in immediate downvotes. The whole karma system helps with trustworthiness.
- Most subs are moderated well and spam gets removed quite quickly
That being said, good fake reviews are technically almost impossible to detect, even with sophisticated network analysis of the reviewer's profile.
Looria is still in Beta and we're working on improving our classification, summarization, and sentiment analysis. Let me know what you think!