Reverse Dictionary Search

Feedback and Developments
1

I want to search for a definition and find the matching words.

When I do this search on DuckDuckGo, I am able to see my exact terms in each snippet and in bold.

"incapable of" site:merriam-webster.com/dictionary

When I do this search in Mojeek, the exact phrase is not shown in each snippet. This makes it difficult to know if the results are relevant without visiting each result.

"incapable of" site:merriam-webster.com inurl:dictionary

Examples

Here are some examples where the results and snippets are excellent. This time, I am searching for information about a book quote.

"direction we are moving" site:http://conwayyao.com/civilization-v-quotes/

"direction we are moving" intitle:"oliver wendell holmes" site:gutenberg.org
  • Video Game Quote: One relevant result with my search terms centered in the snippet.
  • Book Quote: Two results including a snippet from the full-text book.

An exact match can also save time even when the snippet does not cooperate.

Here I remembered an unusual detail from this documentation. And I performed the same steps as a reverse dictionary search.

"Wolfgang Amadeus Mozart" site:microsoft.com/en-us/powershell
  • Mozart Mention: One result from the documentation but an irrelevant snippet.

Feature Request

The key features of the reverse dictionary search method are:

  • Search for an exact match.
  • Search a specific site including subdomains and paths.
  • Display my search terms in the snippet.

With respect to Mojeek, please:

  • Show no results if I ask for an exact match and there are none.
  • Consider combining site: and inurl: for simplicity and specificity.
  • Show the exact match in each snippet.

References

Reverse Dictionary | Wikipedia

Search Operators | DuckDuckGo

Search Operators | Mojeek

Subdomains | Wikipedia

URL Paths | Wikipedia

3 Likes
2

Hey @mike, cheers for this, we’re working on a reply that answers your request here, with a detail that matches it in terms of the amount of thought put in. I will hopefully have this for you tomorrow.

1 Like
3

The response, a day late (apologies), we thought it best here to address the feature requests directly; thanks for the examples though, frequently we get requests in that don’t have this context and, although any feedback is appreciated, feedback with examples is way more useful.

Prefacing this with something that you already know, DDG is pulling their web results from Bing, who have more compute power than us to generate the snippets, which can be an intensive task. Because of that, we’ve probably not gone through thoroughly enough the examples that you’ve used.

On:
- Search for an exact match.
- Display my search terms in the snippet.

The search you’re doing here is matching exactly, but the exact match is not displaying in the snippet in bold as with DDG. That is definitely something that we can see as causing problems, but if you navigate to the pages linked you should be able to find that exact text on all of them; in fact, we have a short message that pops up in order to inform people that they have reached the end of the list of exact matches which you can see here at the top:

https://www.mojeek.com/search?q=%22no+exact+phrase+match+found%22

There were no exact matches for “no exact phrase match found”

or here in roughly the middle of the page:

https://www.mojeek.com/search?q=%22incapable+of%22+site%3Amerriam-webster.com+inurl%3Adictionary&s=121&rbb=0

This being said, the lack of that bolded text-in snippet is definitely confusing and it’s something we will look into.

On the subdomains and paths:

Subdomains do work, i.e:

https://www.mojeek.com/search?q=mojeek+site%3Ablog.mojeek.com&rbb=0

will just produce results under blog. whereas:

https://www.mojeek.com/search?q=mojeek+site%3Amojeek.com&rbb=0

will product results from mojeek.com and blog.mojeek.com and any other subdomains. Paths is a bit more difficult and not something that we’re currently able to offer; implementation would not be straightforward, but that is not to say it can never happen.

On consider combining site: and inurl: for simplicity and specificity:

This is possible, but due to the way that Mojeek is set up it wouldn’t work perfectly. We can look into this when we have time to see if it is something we could roll out in the future.

As with up top, this level of detailed feedback is extremely useful and so thanks once again.

2 Likes
4

Thanks for this. It is helpful to see how Mojeek relates to my example.

It looks like I can get the same results searching with Mojeek. But I just need to click through which is fine.

One detail about the URL is

site:merriam-webster.com/dictionary

can only pattern match that string whereas

site:merriam-webster.com inurl:dictionary

might match

https://www.merriam-webster.com/dictionary/dictionary

and other combinations of path which might be unhelpful.

This mismatch is part of what motivated me to mention site:

1 Like
5

I have another example for the site operator / inurl operator split.

I am searching for the weather forecast from the local television affiliate.

The old website was cbs2chicago.com So, I could perform a search using the site operator.

# Query
site:cbs2chicago.com weather

# URL
https://www.mojeek.com/search?q=site%3Acbs2chicago.com+weather

# Decoded URL
https://www.mojeek.com/search?q=site:cbs2chicago.com+weather

The new site is cbsnews.com/chicago

I can get the correct result site:www.cbsnews.com cbsnews chicago weather

But, I can’t perform my intended search site:cbs2chicago.com/chicago weather

1 Like
6

Thanks a lot @mike for the updated example, a credit to yourself for remembering something from a while back as needing more examples - I’ve added it into that issue. All it took was seemingly a large web content centralization project on the part of CBS :smile:

1 Like
7

I didn’t remember this. Instead, I thought I wrote about the intent thing before. And, this thread cropped up in my forum searches. So, I moved that part of my draft here. :slightly_smiling_face: I was going to put my comments in a feedback form. But I needed Markdown formatting. My plain text draft was unintelligible.

1 Like
8

Considering that Mojeek can already run AI summaries which I’ve read are equally computationally intensive, can this problem regarding snippets finally be resolved?

Also, I see a potential for snippets to be superior to AI summaries and chatbots, while at the same time maintaining Mojeek’s advantage on organic hyperlinks. In my PC’s browser, I tried expanding the snippets to around 4 lines (450 characters), and I find it useful to get a gist of each result, for identifying those possibly relevant, and eliminating off-topic ones (with an additional benefit of giving a wide variety of terms for filtering the results using search operators). It seems that quoting the articles directly is much more reliable than the hallucinating AI.

1 Like
9

This bit from me is from February of 2022, since then we have been working on making snippet/summary generation less expensive/intensive and there are further improvements to come for sure.

2 Likes
10

@mike we’ve shipped an improvement which was targeted at the issue pointed out here, but it has the ability to improve a bundle of other snippets. If you had any other examples where you’d found similar problems then I’d be very eager to hear what happens when running them through this updated version :pray:

1 Like