Improvements to EPG matching

[InJustIs]

I've been having some issues with EPG matching, and trying to fix them when you have hundreds or thousands of channels is a little difficult because the only two options right now are "Auto set all EPG" and manually setting each channel one by one. Of course the reliability of auto setting EPG depends on the formatting, but a lot of the time I have to rely on opensource EPG which aren't ideal. So I had a few ideas that can help with this and I've ranked them in terms of what I think would be helpful. (1 being most helpful)

1. Provide an option to "assign" a specific EPG to a stream group. ( e.g. I would create a stream group for free online channels, then I would assign "Free-Online-Channels-EPG" to that streamgroup for it to auto set the EPG from). This would make it less likely for other EPG interfering or incorrectly being assigned to channels.
2. Allow a setting to adjust the minimum threshold for the levenshtein distance calculation that's used to auto assign EPG. Anything below that threshold would not get assigned, allowing me to sort the list based on "EPG=None" and assign the missing channels myself manually. This would ideally need recommendation # 3 to work best.
3. Allow bulk adding manual EPG using the channel select function. Right now when I select a channel, the only EPG option for me is to "auto set (1) EPG". It would be great to have an option to "Manually set (1) EPG" so I can select however many channels i want and manually pick out a EPG channel to assign. This might seem intuitive, but I have a lot of backup channel sources and although ideally they should all be combined so they act as backup streams, it's way too time consuming to go through hundreds/thousands of channels to format them so I often just leave them as "Channel", "Channel (Source 2)", "Channel (Source 3)"
4. Add an additional column in the channels section that lists the results of the levenshtein distance so I can sort it by the lowest values and adjust the EPG as needed. This option won't be needed if the second suggestion is implemented.

[carlreid]

A lot of this makes sense, and making this matching easier would certainly be beneficial. I think some of these changes are likely more difficult than others.

1. I guess this would be medium difficulty. Likely needs some DB changes, logic changes, and UI changes
2. Should be easy, then it's just an update to settings and then using that value instead of a pre-defined one.
3. Not sure I understand this one. Would you then manually bulk update a selection of channels to the same EPG value?
4. Might be on the easier side to do, as it would only be output data rather than change any logic.

I've instead dabbled a little with adding a new EPG matching approach, using a model called MiniLML6v2 which is good at sentence similarity. Essentially what is attempting to be done here, is finding the similarity between a Channel Name and an EPG name. If this is something you would be willing to try out, then you can try use the Docker Image of ghcr.io/carlreid/streammaster:pr-106 once the build completes.

From my testing, it seems to do pretty well. Though it currently takes a bit of time to work (a minute or two it was for me) - maybe that's due to some bad code of my own somewhere. Probably makes sense to spin up a new instance of StreamMaster with an unmapped list of channels to try it out, so not to ruin your curated one at the moment.

[InJustIs]

A lot of this makes sense, and making this matching easier would certainly be beneficial. I think some of these changes are likely more difficult than others.

1. I guess this would be medium difficulty. Likely needs some DB changes, logic changes, and UI changes

2. Should be easy, then it's just an update to settings and then using that value instead of a pre-defined one.

3. Not sure I understand this one. Would you then manually bulk update a selection of channels to the same EPG value?

4. Might be on the easier side to do, as it would only be output data rather than change any logic.

I've instead dabbled a little with adding a new EPG matching approach, using a model called MiniLML6v2 which is good at sentence similarity. Essentially what is attempting to be done here, is finding the similarity between a Channel Name and an EPG name. If this is something you would be willing to try out, then you can try use the Docker Image of ghcr.io/carlreid/streammaster:pr-106 once the build completes.

From my testing, it seems to do pretty well. Though it currently takes a bit of time to work (a minute or two it was for me) - maybe that's due to some bad code of my own somewhere. Probably makes sense to spin up a new instance of StreamMaster with an unmapped list of channels to try it out, so not to ruin your curated one at the moment.

I just tried pr-106, and it's definitely a lot better in matching the channels. I think out of the thousand channels or so that I skimmed over, I only caught about 20-30 channels that were wrong. The rest, especially the ones that were previously problematic, matched correctly with the new model. Previously, around 40% of the channels were incorrect so this is a huge improvement.

To clarify suggestion 3, yes you're correct. I would like to select multiple channels and assign the same EPG value. This is purely a result of not consolidating streams, and instead leaving them independent. Either way, with the new model it's not a huge issue since I only had a few channels to correct and it wasn't as much of a hassle.

[carlreid]

Great to hear that it worked well for you 👍

I think I'll then try to see if the "Auto EPG" should give options of using Fuzzy Matching or "AI", maybe also even with the ability to change the accepted threshold for both of those. Then the additional suggestions here could be considered later.