Why is Search Bar So Hit & Miss?

[ruhrohshingo]

iirc MD moved to an Elasticsearch backend for most of the data, which also caused a significant change to how the search itself functions which was maybe 2-3 years ago. Before the switch to Elasticsearch search was already kind of spotty, but I tolerated the flakiness of search on ES for a while hoping it'd be refined to be more reliable and precise. This doesn't seem to be the case (or I'm just searching wrong relative to how MD expects people to search).

My most direct examples are things like titles which are composed of a single, relatively common word. "HITS" is a good example of this where the "must include" syntax in double quotes (if still valid) isn't always respected. Likewise, for titles that aren't necessarily comprised of one or more common "garbage" terms (e.g.: the, or, and, is, be, am, etc.), matches either turn nothing up or unrelated entries a surprising amount of the time. A recent example was I attempted to search for "Abandoned Bastard" leaving off the rest of the title initially (full title is "Abandoned Bastard of the Royal Family") which turned up no results whatsoever for the first two terms and didn't match until I had typed the entire name of the title.

The other oddity and something of an annoyance I've seen is where titles may or may not use "wo" instead of "o" for the particle を/ヲ. In several cases where "o" is used, "wo" will be a complete miss.

I've worked with ES a bit in the past, so I'm aware it has capabilities of taking a search term and identifying with certain weights "matches" which helps account for things like misspelling and grammar in user input. I can understand if getting the correct search results more often is complicated by most modern titles having numerous aliases, stylistic Romanization deviations of things like "desu ga" vs "desuga" and "janai" vs "ja nai", and if the initially used title turns out to not be the officially adopted one and in the cases of extremely long winded titles being shortened.

Finally, scrounging around a bit to see if there's a search guide for MD before posting this turned up nothing that was official or along the lines of what I'm asking about. If there's any info MD can chime in on this - even just a "we're aware, but the fix is really painful" - is appreciated. I'd rather understand why things are the way they are than continue putting up with it begrudgingly wondering why it can't be better. 🙇‍♂️

 

[Roler]

While I agree that MD search is not perfect, it is definitely more refined from what it was initially, and I honestly don't run into any issues most of the time (unless some manga is missing an alt title, but search is not at fault there).

A recent example was I attempted to search for "Abandoned Bastard" leaving off the rest of the title initially (full title is "Abandoned Bastard of the Royal Family") which turned up no results whatsoever for the first two terms and didn't match until I had typed the entire name of the title.

Searching for "Abandoned Bastard" returns the wanted result unless you misspelt it, and in case you did and typed the rest of the title (e.g. "Abndoned Bstad of the Royal Family"), it's going to return the wanted result.

Dunno if search will improve from the points you made, but if you can't find a result no matter what, copy-paste the original title

 

[Jadian]

Something certainly does seem strange here.

Searching for "hits" returns the exact same search results as searching for hits~, despite a fundamentally different expectation given the search syntax.

The ~ in the second example is for fuzzy search, which allows things like "hito" or "hist" count as matches as well.

[edit] From what I can see, enclosing a search phrase in quotation marks applies fuzzy search to each word individually (more mistakes allowed for longer words), can drop words in longer phrases, and does not care about word order.

Hopefully this helps with understanding what's going on with your searches.

 

[Roler]

Something certainly does seem strange here.

Searching for "hits" returns the exact same search results as searching for hits~, despite a fundamentally different expectation given the search syntax.

The ~ in the second example is for fuzzy search, which allows things like "hito" or "hist" count as matches as well.

[edit] From what I can see, enclosing a search phrase in quotation marks applies fuzzy search to each word individually (more mistakes allowed for longer words), can drop words in longer phrases, and does not care about word order.

Hopefully this helps with understanding what's going on with your searches.

As far as I know, there is no search syntax like "hits" or hits~ on MD, you can only search the title normally, and adding those just confuses the search, but I would also like them to be implemented.

Btw this looks more like a suggestion thread than a bug thread 🤔