-
Suggestion
-
Resolution: Unresolved
-
None
-
Confluence 2.5.1 webapp, running inside Jira standalone on Win2003, against SQL-Server database. JDK1.5
-
158
-
26
-
NOTE: This suggestion is for Confluence Server. Using Confluence Cloud? See the corresponding suggestion.
We would like to able to search for exact text, and not have Lucene tokenise, remove stop words, or do any stemming.
We partly use Confluence as a knowledge base for our own software product, which includes lots of system codes, acrnoyms etc. The specific use case that prompted this was a user wanting to search for the text "S/D". Lucene strips off the 'S', and the '/', leaving the search to just look for 'd'. Not particularly useful.
It would be brilliant if we could enter a query somehow that overrode this.
Raised following support issue CSP-29584
Workaround
We've implemented the functionality requested, but are awaiting a point where we can have a reindex performed as part of upgrading Confluence in the future. Until that point, we cannot turn the feature on by default. In the interim you can follow the below steps to enable the functionality in your instance.
You will need the following versions of Confluence
- Confluence 8.0.0+ to do exact searching on the titles of pages and blogs
- Confluence 8.6.0+ (not yet released at time of writing) to do exact searching on the content of pages and blogs
To enable the functionality, you will need to
- Be running the above listed versions of Confluence
- Enable the following dark feature in Confluence confluence.search.improvements.exact
- Rebuild the Search Index Note: This will need to be done again when crossing either the Confluence 8.0.0 or 8.6.0 line
- Run a search using "quotation marks" to indicate an exact search is desired
Be aware that this change will have an impact on the size of the index in your instance, and the time taken to index. From our testing, we saw a 10%-15% increase in both index size and time to index. This time will depend heavily on the shape of the data in your instance.
- is duplicated by
-
CONFSERVER-53365 Phrase search with stop words is extremely broken
-
- Closed
-
-
-
- Closed
-
-
-
- Closed
-
- is related to
-
-
- Long Term Backlog
-
- relates to
-
-
- Short Term Backlog
-
-
- Closed
-
- Gathering Interest
- was cloned as
-
- Gathering Interest
- mentioned in
-
Page Loading...
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
[CONFSERVER-14910] Provide ability to override Lucene tokenisation and stemming and search for exact text (literal search)
Has this "suggestion" really been open for more than 15 years?
Yikes! At least I'm not paying the bill.
Typical.
It takes so long for Atlassian to implement this (15 years!) that by the time they do, my employer has jumped ship to an alternative product.
This is a major problem for our Confluence Data center 8.7.x I cannot separate kb-how-to-articles from kb-troubleshooting-articles at ALL. Also if the page has no label then it shows up in the search results on the portal–regardless of the permission of the page. NOT good for my gov't client.
Hey c09381d4440b,
It does indeed. Let us know how you get on 
Thanks,
James Ponting
Engineering Manager - Confluence Data Center
Hi All,
A quick update on this one.
We've been working on introducing this functionality for a while, and have introduced it in stages with the intent of enabling it all in one piece. The work requires reindexing Confluence, and so we were hoping to release all of the feature alongside some planned work that also required a reindex to complete. Unfortunately that work has fallen off the roadmap, leaving us in a bit of an uncomfortable situation where we have an improvement to ship, but it requires manual intervention to enable. So here's how you do it manually as we work out how we'll roll this out for everyone.
You will need the following versions of Confluence
- Confluence 8.0.0+ to do exact searching on the titles of pages and blogs
- Confluence 8.6.0+ (not yet released at time of writing) to do exact searching on the content of pages and blogs
To enable the functionality, you will need to
- Be running the above listed versions of Confluence
- Enable the following dark feature in Confluence confluence.search.improvements.exact
- Rebuild the Search Index Note: This will need to be done again when crossing either the Confluence 8.0.0 or 8.6.0 line
- Run a search using "quotation marks" to indicate an exact search is desired
Be aware that this change will have an impact on the size of the index in your instance, and the time taken to index. From our testing, we saw a 10%-15% increase in both index size and time to index. This time will depend heavily on the shape of the data in your instance.
Hopefully this will be of some help to you all. I'll update the ticket when we properly release the feature to the broader public (i.e. without the need for a dark feature), but this just depends on when we can organise an upgrade reindex.
Thanks,
James Ponting
Engineering Manager - Confluence Data Center
This impoverished searching ability makes the tool less desirable by teams that use more "primitive" tools that support it. Harder to sell. Seems like this should be slam-dunk.
Hey Colin, would love to hear more about this ChatGPT feature plugin. My organization is looking for an AI Chatbot like ChatGpt to be added to the confluence search so that it gives the user specific results pertaining to the search and not irrelevant stuff. So, I don't know whether Atlassian is working on that or not but I need to know. Can we connect if it is not too much trouble?
We really look for a solution for this. Seems like there is a new plugin using openAI chatGPT like, is anyone used it? I was not sure if that's might help I have installed it on different account for testing, please let me know what you think..
I don't know anything about Confluence indexing so I asked a 'friend', will be useful as a 14th birthday present:
Me:
Can you create a lucene index tokenising configuration that will use a string with multiple period characters as a single token but still use period at the and of a word as a stop character? So for example 127.0.0.1 will be tokenized as 127.0.0.1 but stop. will be tokenised as stop
Yes, you can achieve this by using a custom tokenizer and adding a specific pattern to tokenize the string with multiple periods as a single token.
Here is an example configuration for Lucene using the StandardTokenizer with a custom pattern for the string with multiple periods:
code
{{<tokenizer class="org.apache.lucene.analysis.standard.StandardTokenizerFactory"/>
<tokenizer class="solr.PatternTokenizerFactory" pattern="([\d][.]){3}[\d]|[^\s\.]+"/>
<filter class="org.apache.lucene.analysis.standard.StandardFilterFactory"/>
<filter class="org.apache.lucene.analysis.LowerCaseFilterFactory"/>
<filter class="org.apache.lucene.analysis.StopFilterFactory" ignoreCase="true" words="stopwords.txt"/>
<filter class="org.apache.lucene.analysis.PorterStemFilterFactory"/>}}
In this configuration, the StandardTokenizer is used to tokenize the input text into individual words. Then, the PatternTokenizer is used to tokenize the string with multiple periods as a single token using the regular expression pattern ([\d][.]){3}[\d].
The | symbol is used to separate multiple patterns, so the second part of the pattern [^\s\.]+ will match any non-space, non-period characters and treat them as separate tokens.
Finally, the StandardFilterFactory is used to remove any punctuation and special characters, followed by the LowerCaseFilterFactory to lowercase all the tokens, the StopFilterFactory to remove any stop words and the PorterStemFilterFactory to apply stemming to the tokens.
Note that you can adjust the filter chain to suit your specific requirements.
Me:
Can you make PatternTokenizerFactory pattern more generic so it includes any string with multiple period characters inbetween alphanumeric characters not case sensitive
Certainly! Here's an updated pattern that will match any string with multiple period characters between alphanumeric characters, and it will be case-insensitive:
code
<tokenizer class="solr.PatternTokenizerFactory" pattern="(?i)\b(\w+\.)\w\b|\w+"/>
In this pattern, the (?i) flag is used to indicate that the pattern should be case-insensitive. The \b characters denote word boundaries, ensuring that the pattern only matches full words.
The \w+ pattern matches one or more alphanumeric characters, while \. matches one or more periods. Together, (\w\.) matches one or more groups of one or more alphanumeric characters followed by a period. The final \w matches one or more alphanumeric characters after the last group of periods.
The | character separates the two parts of the pattern. The second part, \w+, matches any single word containing only alphanumeric characters (no periods).
This pattern should correctly tokenize any string with multiple period characters between alphanumeric characters, while still treating periods at the end of a word as stop characters.
Me:
Can you test the regex pattern against finally.i.can.search.for.ip
Sure! Let's test the updated pattern against the string "finally.i.can.search.for.ip":
code
{{Pattern pattern = Pattern.compile("\\b(\\w+\\.)\\w\\b|
w+");String input = "finally.i.can.search.for.ip";Matcher matcher = pattern.matcher(input);while (matcher.find())
}}
The output of this code should be:
code
finally.i.can.search.for.ip
This indicates that the input string has been successfully tokenized into a single token, as desired.
Note that the regular expression uses double backslashes to escape special characters (such as \b and \w) in the Java code.
How could they fix this when they were busy making sure we have new emojis to use in the latest version of Confluence. Emojis are a much bigger value add.
The rampant stemming is hindering more than it is helping in most people's use-cases.
Get ready to <sarcasm>celebrate</sarcasm> this ticket's 14th birthday on 13th March.
Why does a search for "searchable" return results for "search"?
Is "able" a stop word now?
This would be a huge improvement for the search!
People tend to search for content by using the search box. Yes, we can filter, but the better way is to refine the search logic, suitable for all users.
+1 on this issue. double quotes is a standard for exact matching, without this the search is useless.
> define exact
exact
adjective
- not approximated in any way; precise
please rename this feature to 'approximate match' if you aren't going to implement exact match, thanks
Current search is optimized for natural language.
But the real-world use for many of us is that Confluence is a technical reference, where every symbol and letter is significant. Please give an option for exact-match!
We need to be very structured in how the wiki is used, everything needs a label and labels should be obligatory. Lots wrong with this out of the box and the search is just another one.
Because of this issue, using Confluence is like using the internet before Google existed.
Please fix this issue, within my company I cannot search for the exact match of an acronym that ends with an 'S'. This is causing a headache because there are two different acronyms that are identical save for the 'S' character at the end. If I search for an exact word via quotation marks, I want the word or phrase exactly as it is written. Additionally, I noticed that the 'S' character is uniquely removed from the phrase, no other character exhibits this behavior, and it only occurs on words that are three characters or longer. I don't understand how this issue can go unaddressed for thirteen years.
The have it open so long because it doesn't show up in their search. Search in Confluence is pitiful. Any improvement, this included would be a good one.
13 years later... >= 13 related bugs Closed as "Duplicates" of this suggestion.
Get your stuff together, atlASSian (emphasis intended)
Still nothing on this I see. How many people have to request this before they add it 
Edit: I found a trick that helps with this at least, if you are searching for a single word you can append * to it to force wildcard matching to take over and disregard the stemming behavior.
Atlassian, listen up! What exactly is the point of leaving this fundamental ticket open for 13 YEARS as a "Future Consideration". Just close it and be proud of your lack of caring for your users.
If this was open source this obvious defect would have been fixed 13 years ago, but we pay for your product and get nothing here.
....'don't touch anything within quotes'........'don't touch anything within quotes'....
That's the echo of your request, bouncing back from a conversation completely empty of any response or action from Atlassian!
I'm cringing with vicarious embarrassment. It's a no-brainer. And would probably take an hour to code and test 'don't touch anything within quotes'.
Wow - this is incredible, what a disservice to your users. An insane feature to remain unimplemented for 13 years.
13 years. Unlucky for us and every user that just wants a usable search function.
Happy Birthday CONFSERVER-14910!
13 years old! We better start planning on how to finance your college education!
To work around this bug, I generate a PDF of the page family (or the whole wiki space) and then search in that PDF. I agree that we shouldn't have to do this, but ... .
Can anyone recommend a Confluence replacement where the search function works for quoted phrases?
search for exact text ist absolutely essential for working with gigantic collection of pages - cannot understand why this feature ist not available in confluence. Dont explain it, implement it! - Thanks in advance!
Searching in ADLM tools is not intuitive. As an end user, I want to be able to enter "what I am looking for" and be able to find those stories that contain the phrase, wherever it appears in the story.
I don't know why this isn't just fixed: quotes mean "take everything inside the quotes as-is".
The reason why we use quotes is to disable any special treatment (OF COURSE that should include stopwords, hyphens,....).
So this is not a "feature requests". It is a fundamental bug that can't be fixed fast enough....
Arrived here via the obvious search for confluence+search+exact+phrase, having found that searching for our project's git name is useless (since it's the project name concatenated with hyphens). I guess it becomes a rite of initiation: "why does your repository have a gibberish name? because that way we can search for it in Confluence"; would be better if the tool fits the human workflow & not the converse.
I arrived here because searching our Confluence server for "definition of done" fails to match anything, even though the search without the quotes shows there are dozens of matches. Unfortunately, the quote-less search turns up 2900 hits, so hit quality is unusably low.
Please modernize ur search algorithm a better time for everyone involved
12y and 600 customers actively asking for it + most opensource freeware I use has a better search algo
imagine
There are 50+ tickets for CONFSERVER in the FUTURE CONSIDERATION status.
It can take years until Atlassian does something.
The only possibility of acceleration that I see is to get more votes.
I'm actively trying to find more people that will vote for this issue.
We all have to do this to get the party started…
I have been burnt by this. This needs to be fixed, I wish I knew about this before I suggested Confluence as our company pages. Searching for an exact string should be the default, its impossible to search for domains, error strings, strings with dots or other characters.
+1, I don't need to add a lot. It is a feature which we also miss urgently. We only recently decided to use Confluence in our company and for our customers. I just did not expect that the search feature of Confluence does not have an exact match and is quiet limited for a mature platform as it is. This gives us headaches and in my opinion is not up to date.
I'd say, given Confluence will be a cloud only product soon, most of us don't care about this feature anymore as for many people a cloud based Wiki solution to manage internal company knowledge is not a viable solution.
Given this is a 12 year old ticket, and somehow still never made it into the product, perhaps a better approach for all of us would be to just write our own search tool that can be pointed to a wiki and does what makes sense rather than what Confluence product management thinks.
It would be nice to have "exact searches" exact, how is that so hard to understand if anyone at Atlassian is out there please respond and provide some clarity on this.
Please add this feature!
We document our software's "exact error messages" in confluence so that we can search for them later to find the specific resolution to the issue. If the exact "double-quoted" error message we're searching contains a stop-word, it excludes the entire page from the search result! This is very frustrating since there is no method around this limitation since (since full "non-quoted" error message returns far to many results to be helpful).
I want to search for the error message "repository not found", what do i get "repository found". That isn't really helpful :\
Search text in quotes should only match that exact phrase. If I'm searching for "auditing", I don't want results with audit, just auditing. To say that doing so is a "Search for an exact match" and you don't get an exact match, is just wrong.
Imo, one of the most important features of a wiki is the search.
Despite its more "enterprisy" of approach than just a graph of linked pages, search is still the most prominent way on how pages are accessed for us.
Being able to search for the terms given the specific glossary of the company is hugely important and not supporting this is arguably inconsistent with what (at least I understand) is the vision of the product.
Need an exact search function. The current implementation is severely limited especially if you have uniform file naming.
It's misleading and frustrating to offer search by an "exact phrase" except it's not due to your implementation. I have a lot invested in Confluence and not having a search tool to work it is very frustrating.
Yes it has.
Welcome to the party.