Viktor Lofgren
099133bdbc
(index) Fix verbatim match score after moving full phrase group to a separate entity
2024-08-25 10:43:35 +02:00
Viktor Lofgren
b09e2dbeb7
(build) Fix dependency churn from testcontainers
...
Apparently you need to pull in commons-codec now in order to run testcontainers, through spooky action at a distance.
2024-08-25 10:35:48 +02:00
Viktor Lofgren
96bcf03ad5
(index) Address broken tests
...
They are still broken, but less so.
2024-08-25 10:34:36 +02:00
Viktor Lofgren
0999f07320
(search-query) Add new ranking parameters for proximity and verbatim matches
2024-08-25 10:34:12 +02:00
Viktor Lofgren
9eb1f120fc
(index) Repair positions bitmask for search result presentation
2024-08-22 11:28:23 +02:00
Viktor Lofgren
b0a874a842
(*) Upgrade slop library -> 0.0.5
2024-08-18 11:05:27 +02:00
Viktor Lofgren
93652e0937
(qdebug) Accurately display positions when intersecting with spans
2024-08-15 11:55:48 +02:00
Viktor Lofgren
0a383a712d
(qdebug) Accurately display positions when intersecting with spans
2024-08-15 11:44:17 +02:00
Viktor Lofgren
03d5dec24c
(*) Refactor termCoherences and rename them to phrase constraints.
2024-08-15 11:02:19 +02:00
Viktor Lofgren
a18edad04c
(index) Remove stopword list from converter
...
We want to index all words in the document, stopword handling is moved to the index where we change the semantics to elide inclusion checks in query construction for a very short list of words tentatively hard-coded in SearchTerms.
2024-08-15 09:36:50 +02:00
Viktor Lofgren
92522e8d97
(index) Attenuate bm25 score based on query length
2024-08-15 08:41:38 +02:00
Viktor Lofgren
049d94ce31
(index) Add body position match to qdebug fields
2024-08-15 08:39:37 +02:00
Viktor Lofgren
dbc6a95276
(index) Consume the new 'body' span in index to make it used in ranking
2024-08-15 08:33:43 +02:00
Viktor Lofgren
75b0888032
(slop) Migrate to latest Slop version
2024-08-14 11:44:35 +02:00
Viktor Lofgren
623ee5570f
(slop) Break slop out into its own repository
2024-08-13 09:50:05 +02:00
Viktor Lofgren
e6c8a6febe
(index) Add index-side deduplication in selectBestResults
2024-08-10 10:51:59 +02:00
Viktor Lofgren
4ece5f847b
(index) Add more qdebug factors
2024-08-10 10:45:30 +02:00
Viktor Lofgren
e4f04af044
(index) Give BODY matches a verbatim match value
2024-08-10 10:22:19 +02:00
Viktor Lofgren
b730b17f52
(index) Correct handling of firstPosition to avoid d/z
2024-08-10 10:21:59 +02:00
Viktor Lofgren
98c40958ab
(index) Simplify verbatim match calculation
2024-08-10 09:54:56 +02:00
Viktor Lofgren
41b52f5bcd
(index) Simplify verbatim match calculation
2024-08-10 09:51:03 +02:00
Viktor Lofgren
016a4c62e1
(index) Bugs and error fixes, chasing and fixing mystery results that did not contain all relevant keywords
2024-08-10 09:51:03 +02:00
Viktor Lofgren
df89661ed2
(index) In SearchResultItem, populate combinedId with combinedId and not its ranking-removed documentId cousin
2024-08-09 16:32:32 +02:00
Viktor Lofgren
2e89b55593
(wip) Repair qdebug utility and show new ranking details
2024-08-09 12:57:25 +02:00
Viktor Lofgren
7babdb87d5
(index) Remove intermediate models
2024-08-07 10:10:44 +02:00
Viktor Lofgren
f01267bc6b
(index) Don't load fwd index offsets into a hash table at start.
...
This makes the service take forever to start up. Memory map the data instead and binary search. This is a bit slower, but not by much.
2024-08-06 11:16:28 +02:00
Viktor Lofgren
df6a05b9a7
(index) Avoid hypothetical divide-by-zero in tcfAvgDist
2024-08-06 10:55:57 +02:00
Viktor Lofgren
8569bb8e11
(index) Avoid divide-by-zero when minDist returns 0
2024-08-06 10:34:05 +02:00
Viktor Lofgren
ca6e2db2b9
(index) Include external link texts in verbatim score
2024-08-06 10:23:23 +02:00
Viktor Lofgren
2080e31616
(converter) Store link text positions
...
To help offer verbatim matches for external link texts, we assign these positions in the document a bit after the actual document ends. Integrating this information with the ranking is not performed here.
2024-08-04 12:00:29 +02:00
Viktor Lofgren
ee49c01d86
(index) Tune ranking for verbatim matches in the title, rewarding shorter titles
2024-08-03 14:47:23 +02:00
Viktor Lofgren
b21f8538a8
(index) Tune ranking for verbatim matches in the title, rewarding shorter titles
2024-08-03 14:41:38 +02:00
Viktor Lofgren
dd15676d33
(index) Tune ranking for verbatim matches in the title, rewarding shorter titles
2024-08-03 14:18:04 +02:00
Viktor Lofgren
ec5a17ad13
(index) Tune ranking for verbatim matches in the title, rewarding shorter titles
2024-08-03 14:07:02 +02:00
Viktor Lofgren
8462e88b8f
(index) Add min-dist factor and adjust rankings
2024-08-03 13:07:00 +02:00
Viktor Lofgren
bf26ead010
(index) Remove hasPrioTerm check as we should sort this out in ranking
2024-08-03 13:06:50 +02:00
Viktor Lofgren
c2cedfa83c
(index) Experimental ranking signals
2024-08-03 10:33:41 +02:00
Viktor Lofgren
c6c8b059bf
(index) Return some variant of the previously removed 'Bm25PrioGraphVisitor'
2024-08-03 10:10:12 +02:00
Viktor Lofgren
d8a99784e5
(index) Adding a few experimental relevance signals
2024-08-02 20:26:07 +02:00
Viktor Lofgren
e2107901ec
(index) Add span information for anchor tags, tweak ranking params
2024-08-01 11:46:30 +02:00
Viktor Lofgren
15745b692e
(index) Coherences need to be able to deal with null values among positions
2024-07-31 22:00:14 +02:00
Viktor Lofgren
dc5c668940
(index) Re-enable parallelization of index construction, disable parallel sorting during construction
...
The first change, running index construction in parallel, was previously how it was done, but it was changed to run sequentially to see how it would affect performance. It got worse, so the change is reverted.
Though it's been noted that sorting in parallel is likely not a good idea as it leads to a lot of I/O thrashing, so this is changed to be done sequentially.
2024-07-31 10:06:53 +02:00
Viktor Lofgren
b316b55be9
(index) Experimental initial integration of document spans into index
2024-07-30 12:01:53 +02:00
Viktor Lofgren
34703da144
(slop) Support for nested array types and array-of-object types
...
Also adding very basic support for filtered reads via SlopTable. This is probably not a final design.
2024-07-29 14:00:43 +02:00
Viktor Lofgren
1282f78bc5
(slop-models) Fix incorrect column grouping leading to errors in converter
2024-07-29 11:01:18 +02:00
Viktor Lofgren
2d5d965f7f
(slop-models) Fix incorrect column grouping leading to errors in converter
2024-07-29 10:34:33 +02:00
Viktor Lofgren
e585116dab
(slop) Add 32 bit read method for Varint along with the old 64 bit version
2024-07-28 13:20:18 +02:00
Viktor Lofgren
d05a2e57e9
(index-forward) Spans Writer should not be in the index page loop context
2024-07-27 15:17:04 +02:00
Viktor Lofgren
6c3abff664
(slop) Move GCS Slop column to the coded-sequence package
...
This lets the slop library be stand-alone without dependence on coded-sequence.
The change also gets rid of the vestigial seek() method in ColumnReader.
2024-07-27 13:58:45 +02:00
Viktor Lofgren
dcb43a3308
(slop) Introduce table concept to keep track of positions and simplify closing
...
The most common error when dealing with Slop columns is that they can fall out of sync with each other if the programmer accidentally does a conditional read and forgets to skip.
The second most common error is forgetting to close one of the columns in a reader or writer.
To deal with both cases, a new class SlopTable is added that keeps track of the lifecycle of all slop columns and performs a check when closing them that they are in sync.
2024-07-27 13:47:47 +02:00
