oneryalcin commited on
Commit
293a5b3
·
verified ·
1 Parent(s): a27a16a

Add reproducibility dataset artifacts

Browse files
README.md ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pretty_name: Luxical Query Expansion Financial/ESG Dataset
3
+ tags:
4
+ - information-retrieval
5
+ - query-expansion
6
+ - finance
7
+ - esg
8
+ - benchmarking
9
+ - reproducibility
10
+ license: mit
11
+ language:
12
+ - en
13
+ size_categories:
14
+ - 100K<n<1M
15
+ ---
16
+
17
+ # Luxical Query Expansion Financial/ESG Dataset (Preliminary)
18
+
19
+ This dataset package is designed to reproduce the experiments in this repository.
20
+
21
+ ## Included Files
22
+ - `finance_priors_v2.tsv`: Mined term priors from local financial corpus.
23
+ - `queries_eval_finance_esg.json`: 12-query weak-label benchmark set.
24
+ - `quality_benchmark.json`: Policy quality benchmark outputs.
25
+ - `latency_benchmark_combined.json`: Combined-policy latency benchmark.
26
+ - `latency_benchmark_conservative.json`: Conservative-policy latency benchmark.
27
+ - `example_expansions_combined.txt`: Sample expansion outputs.
28
+
29
+ ## Notes
30
+ This is a preliminary research artifact. Weak-label query benchmarks are provided for rapid iteration and should not be treated as a substitute for expert relevance judgments.
example_expansions_combined.txt ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Policy: combined
2
+ Queries: 4
3
+ ---
4
+ Query: adjusted ebitda margin guidance
5
+ - adjusted ebit raw=0.8432 sim=0.9984 prior=0.5394 lex=0.9200 final=0.9447 es_boost=4.779 bucket=profitability
6
+ - adjusted ebitda raw=0.8936 sim=0.9994 prior=0.4736 lex=0.9200 final=0.9389 es_boost=4.755 bucket=profitability
7
+ - operating margin raw=0.7557 sim=0.9915 prior=0.4431 lex=0.9200 final=0.9295 es_boost=4.718 bucket=profitability
8
+ - margin was raw=0.7434 sim=0.9893 prior=0.1249 lex=0.8200 final=0.8859 es_boost=4.544 bucket=profitability
9
+ - wide margin raw=0.7020 sim=0.9766 prior=0.1249 lex=0.9200 final=0.8858 es_boost=4.543 bucket=profitability
10
+ - operating within raw=0.6677 sim=0.9558 prior=0.0000 lex=0.9200 final=0.8566 es_boost=4.427 bucket=profitability
11
+ ---
12
+ Query: allowance for credit losses
13
+ - extra credit raw=0.7694 sim=0.9958 prior=0.0000 lex=0.9200 final=0.8886 es_boost=4.554 bucket=general
14
+ - financial losses raw=0.7616 sim=0.9951 prior=0.0000 lex=0.9200 final=0.8881 es_boost=4.552 bucket=general
15
+ - full credit raw=0.7513 sim=0.9940 prior=0.0000 lex=0.9200 final=0.8872 es_boost=4.549 bucket=general
16
+ - much credit raw=0.7404 sim=0.9926 prior=0.0000 lex=0.9200 final=0.8861 es_boost=4.544 bucket=general
17
+ - losses were raw=0.7365 sim=0.9920 prior=0.0000 lex=0.9200 final=0.8856 es_boost=4.543 bucket=credit_risk
18
+ - credit where raw=0.7312 sim=0.9912 prior=0.0000 lex=0.9200 final=0.8850 es_boost=4.540 bucket=general
19
+ ---
20
+ Query: scope 1 scope 2 emissions target
21
+ - scoped raw=0.8347 sim=0.9987 prior=0.0000 lex=1.0000 final=0.8989 es_boost=4.596 bucket=general
22
+ - full scope raw=0.8490 sim=0.9990 prior=0.0000 lex=0.9200 final=0.8912 es_boost=4.565 bucket=general
23
+ - target specific raw=0.7333 sim=0.9900 prior=0.0000 lex=0.9200 final=0.8840 es_boost=4.536 bucket=guidance_outlook
24
+ - target date raw=0.7291 sim=0.9891 prior=0.0000 lex=0.9200 final=0.8833 es_boost=4.533 bucket=guidance_outlook
25
+ - easy target raw=0.6936 sim=0.9781 prior=0.0000 lex=0.9200 final=0.8745 es_boost=4.498 bucket=general
26
+ - target customers raw=0.6833 sim=0.9733 prior=0.0000 lex=0.9200 final=0.8706 es_boost=4.483 bucket=guidance_outlook
27
+ ---
28
+ Query: capital allocation buyback dividend
29
+ - capital investments raw=0.6680 sim=0.9572 prior=0.1366 lex=0.9200 final=0.8715 es_boost=4.486 bucket=liquidity_leverage
30
+ - diluted share raw=0.6658 sim=0.9555 prior=0.7190 lex=0.9200 final=0.8708 es_boost=4.483 bucket=profitability
31
+ - capital gain raw=0.6934 sim=0.9733 prior=0.0000 lex=0.9200 final=0.8706 es_boost=4.482 bucket=liquidity_leverage
32
+ - equity loans raw=0.6660 sim=0.9557 prior=0.6366 lex=0.9200 final=0.8693 es_boost=4.477 bucket=credit_risk
33
+ - cash management raw=0.6823 sim=0.9671 prior=0.0000 lex=0.9200 final=0.8657 es_boost=4.463 bucket=liquidity_leverage
34
+ - term capital raw=0.6723 sim=0.9605 prior=0.0000 lex=0.9200 final=0.8604 es_boost=4.442 bucket=liquidity_leverage
35
+ ---
finance_priors_v2.tsv ADDED
The diff for this file is too large to render. See raw diff
 
latency_benchmark_combined.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "policy": "combined",
3
+ "num_queries": 12,
4
+ "cold_start_ms": 10472.679375001462,
5
+ "batch_ms": {
6
+ "mean": 188.35713339503855,
7
+ "p50": 184.52537499251775,
8
+ "p95": 202.6674750028178,
9
+ "min": 179.7784169903025,
10
+ "max": 206.5042500034906
11
+ },
12
+ "per_query_ms": {
13
+ "mean": 15.696427782919878,
14
+ "p50": 15.377114582709813,
15
+ "p95": 16.888956250234816,
16
+ "min": 14.981534749191875,
17
+ "max": 17.208687500290882
18
+ }
19
+ }
latency_benchmark_conservative.json ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "policy": "conservative",
3
+ "num_queries": 12,
4
+ "cold_start_ms": 10541.75195802236,
5
+ "batch_ms": {
6
+ "mean": 193.0920831975527,
7
+ "p50": 195.24554099189118,
8
+ "p95": 195.52931661019102,
9
+ "min": 188.8506670075003,
10
+ "max": 195.59312501223758
11
+ },
12
+ "per_query_ms": {
13
+ "mean": 16.091006933129393,
14
+ "p50": 16.270461749324266,
15
+ "p95": 16.294109717515916,
16
+ "min": 15.73755558395836,
17
+ "max": 16.29942708435313
18
+ }
19
+ }
quality_benchmark.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "baseline": {
3
+ "precision": 0.90625,
4
+ "hit_rate": 0.9166666666666666,
5
+ "avg_terms": 7.083333333333333,
6
+ "bad_top1": 0.16666666666666666
7
+ },
8
+ "agreement": {
9
+ "precision": 0.84375,
10
+ "hit_rate": 0.9166666666666666,
11
+ "avg_terms": 6.666666666666667,
12
+ "bad_top1": 0.25
13
+ },
14
+ "intent": {
15
+ "precision": 0.84375,
16
+ "hit_rate": 0.9166666666666666,
17
+ "avg_terms": 7.25,
18
+ "bad_top1": 0.16666666666666666
19
+ },
20
+ "combined": {
21
+ "precision": 0.8333333333333334,
22
+ "hit_rate": 0.9166666666666666,
23
+ "avg_terms": 6.666666666666667,
24
+ "bad_top1": 0.16666666666666666
25
+ },
26
+ "conservative": {
27
+ "precision": 0.8645833333333334,
28
+ "hit_rate": 0.9166666666666666,
29
+ "avg_terms": 3.4166666666666665,
30
+ "bad_top1": 0.16666666666666666
31
+ }
32
+ }
queries_eval_finance_esg.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [
2
+ {"query":"adjusted ebitda margin guidance","positives":["adjusted","ebitda","ebit","margin","guidance","operating","profit"],"negatives":["river","scope"]},
3
+ {"query":"net interest margin expansion","positives":["interest","margin","net","yield","nim","banking"],"negatives":["river","scope"]},
4
+ {"query":"allowance for credit losses","positives":["allowance","credit","losses","provision","chargeoff","delinquency"],"negatives":["river","scope"]},
5
+ {"query":"scope 1 scope 2 emissions target","positives":["scope","emissions","target","climate","carbon","decarbonization"],"negatives":["credit","loan","bank"]},
6
+ {"query":"sustainability linked bond","positives":["sustainability","linked","bond","green","esg","climate"],"negatives":["credit","loan","bank"]},
7
+ {"query":"free cash flow conversion","positives":["free","cash","flow","conversion","operating","fcf"],"negatives":["river","scope"]},
8
+ {"query":"same store sales growth","positives":["same","store","sales","growth","comparable","retail"],"negatives":["river","scope"]},
9
+ {"query":"capital allocation buyback dividend","positives":["capital","allocation","buyback","dividend","repurchase"],"negatives":["river","scope"]},
10
+ {"query":"decarbonization roadmap","positives":["decarbonization","roadmap","carbon","emissions","climate"],"negatives":["credit","loan","bank"]},
11
+ {"query":"material weakness internal controls","positives":["material","weakness","internal","controls","remediation","audit"],"negatives":["river","scope"]},
12
+ {"query":"working capital headwind","positives":["working","capital","headwind","cash","inventory"],"negatives":["river","scope"]},
13
+ {"query":"order backlog visibility","positives":["order","backlog","visibility","pipeline","demand"],"negatives":["river","scope"]}
14
+ ]