
Okapi BM25
28
MENTIONS
14
EPISODES
12
PODCASTS
Search complete. 28 mentions across 14 episodes found for "Okapi BM25".
Nov 18, 2026
The King of AEO on Indexing, Podcasts and Entities
J
45:34James DooleyGUEST
But in local, it might work great.
J
45:36James DooleyGUEST
But then you've got also in local, mainly it's query string matching or what's known as being BM25.
J
45:44James DooleyGUEST
which is the quantitative algorithms, which means it's very simple.
J
45:48James DooleyGUEST
It's old school methodology.
J
45:50James DooleyGUEST
It's doing things like correlation.
J
45:53James DooleyGUEST
So if you use things like onpage.ai or marketmoves or surfer SEO, in these BM25 algorithms, they work great because you just get the right engrams, get the right entities, get the right attributes on a page, almost TFIDF.
J
46:11James DooleyGUEST
old school, like keyword stuffing can still work in a quantitative algorithm.
J
46:18James DooleyGUEST
But the minute you start getting traffic or it's in your money, your life, or it's in a highly competitive market and it moves in and burr and mum and these more expensive algorithms start to come and check to see what your page is about.
Agentic search doesn't find pages, it assembles answers
J
2:46Jeff WangGUEST
How does it differ from a database, just a regular database, like a SQL database or something? Yeah, I mean, I
R
2:51Ryan DonovanHOST
know for a lot of old search, you'd use keyword algorithms like BM25 that process the text and had things like that.
R
2:58Ryan DonovanHOST
But for a regular database, you search on a sort of unique index ID, right?
J
3:04Jeff WangGUEST
Yeah, that's right.
P
Unknown podcast
Azure AI Search Hybrid Retrieval - Simply Explained
Nov 12 · 4 Mentions
M
3:58Mirko PetersHOST
Once you type a query, the system scores how well each document matches.
M
4:02Mirko PetersHOST
That scoring method has a name, BM25, and you can understand it without touching the math.
M
4:07Mirko PetersHOST
BM25 gives more weight to words that appear often inside one document, and less weight to words that appear in almost every document.
M
4:14Mirko PetersHOST
Take the word THE for a second.
M
4:15Mirko PetersHOST
It shows up nearly everywhere, so it tells you almost nothing.
M
4:18Mirko PetersHOST
A rare code like XC4471 shows up once, so it tells you a great deal.
M
4:22Mirko PetersHOST
BM25 captures that instinct automatically, so keyword search stays strong on part numbers, exact names, dates, and specialized terms.
M
4:30Mirko PetersHOST
A small example shows how that plays out.
ZeroEntropy with Ghita Houir Alami - Weaviate Podcast #150!
C
41:59Connor ShortenHOST
So I think the next topic, maybe also talking about re-ranking, is agentic search.
C
42:06Connor ShortenHOST
And yeah, we mentioned earlier kind of the success of Grep with agentic search, BM25, and the argument is kind of like, the agent is doing this iterative learning about the corpus while it's searching.
C
42:16Connor ShortenHOST
And it gets this great feedback from Grepper BM 25 because it has like this no matches found.
C
42:22Connor ShortenHOST
And there's a pretty cool paper that's called the questions gambit.
P
Unknown podcast
EP532: MemFit Fix Context Rot
Nov 4 · 1 Mention
S
3:43speaker_0UNKNOWN
During the storage phase, Memfit indexes every single verbatim interaction using dual signals.
S
3:48speaker_0UNKNOWN
It uses a lexical BM25 index to catch exact keywords or specific dates, and a semantic dense index to capture the underlying meaning.
S
3:56speaker_0UNKNOWN
Now, it does do a tiny bit of background summarization to group episodes into segments, but crucially, those summaries are just pointers back to the raw verbatim text.
S
4:06speaker_0UNKNOWN
The heavy, expensive LLM only gets involved at the very end of phase two, when you actually need it to answer a question.
Stanford CS224N NLP with Deep Learning | Winter 2021 | Lecture 12 - Question Answering
D
82:43Danqi ChenGUEST
And this model kind of worked really well And it can largely outperform the traditional IR retrieval models.
D
82:50Danqi ChenGUEST
If you see this figure, so the blue curve here is a traditional IR approach, like a BM25 approach.
D
82:57Danqi ChenGUEST
And the other curve, the orange curve, is basically training this kind of retrieval using only 1,000 question and answer pairs.
D
83:05Danqi ChenGUEST
So by looking at all these different curves, basically using different number of training examples, so it's actually largely outperforming the traditional IR models.
Stanford XCS224U: NLU I Fantastic Language Models and How to Build Them, Part 2 I Spring 2023
C
17:39Christopher PottsHOST
You'll see over here RAM requirements.
C
17:41Christopher PottsHOST
So BM25, It has no hardware requirements at all.
C
17:45Christopher PottsHOST
You can run that on anything.
C
17:48Christopher PottsHOST
Whereas these models down here that have these really high MRR scores, hugely expensive in terms of compute.
C
18:48Christopher PottsHOST
This is kind of interesting too.
C
18:49Christopher PottsHOST
This is from the same paper.
C
18:51Christopher PottsHOST
Here's BM25.
C
18:53Christopher PottsHOST
costs essentially nothing, but it has very low performance.
Agents Are Overtaking Humans in Search: Why Search Must Be Rebuilt
S
45:48Saahil JainGUEST
One is kind of your standard lexical indices.
S
45:50Saahil JainGUEST
So these ones are indices that essentially are using signals like BM25, as well as other You know, you can think of them as like heuristics, you know, related to keywords and term frequencies, et cetera, that allow you to find the right documents.
S
46:04Saahil JainGUEST
So this is generally referred to as lexical retrieval.
S
46:07Saahil JainGUEST
And then there's something called semantic retrieval, which is based on embeddings.
Kyle Roof Ranked Gibberish #1 on Google. Then Google Deindexed 20 of His Sites.
K
64:17Kyle RoofGUEST
And the reason I like that is because, you know, the guys that put Yandex together were Google engineers and they were trying to approximate Google and as close as they can for the least amount of money.
K
64:28Kyle RoofGUEST
And we found out that they're using BM25.
K
64:31Kyle RoofGUEST
And BM 25 is just TF IDF plus word count.
K
64:34Kyle RoofGUEST
And I've been screaming that for years.
Machine Studying with Jacob X. Li - Weaviate Podcast #148!
J
16:27Jacob X. LiGUEST
Um, we kind of have to search iteratively, and we have to decide which one actually, like when we condition on the search results, which part of the search results is actually going to contribute to the actually downstream research.
J
16:41Jacob X. LiGUEST
So one finding from this mini experiment was that even if you use a very outdated model, the model, and given one of the worst retriever, BM25, Compared to a better model, like a newer model, these two models will pretty much cover the same set of papers.
J
17:07Jacob X. LiGUEST
So basically, if I give it a title and some new content, the model will launch a lot of searches.
J
17:14Jacob X. LiGUEST
And if you go through all the searches trajectories, both models will cover pretty much like a a set of paper that's believed to be useful.
C
18:53Connor ShortenHOST
And then you're saying like, therefore it knows how to search more effectively.
C
18:57Connor ShortenHOST
And I guess just that topic of like, how do you search more effectively? How do agents learn to search more effectively? What does it really mean to like write good search queries? I also find that topic to just be really interesting.
C
19:09Connor ShortenHOST
And I guess the other one talking about like using BM25 as a search tool compared to say maybe vector search or late interaction models.
C
19:20Connor ShortenHOST
I know we're maybe getting a little off topic, but just if you don't mind humoring me a little more, I love this kind of agentic search and just taking down this topic.
4 more episodes mention Okapi BM25.
Create an account to see the whole feed, search across every transcript, and follow the entities you care about.