giskardai/giskard-evaluator · Report for cardiffnlp/twitter-xlm-roberta-base-sentiment-multilingual

Hi Team,

This is a report from Giskard Bot Scan 🐢.

We have identified 8 potential vulnerabilities in your model based on an automated scan.

This automated analysis evaluated the model on the dataset cardiffnlp/tweet_sentiment_multilingual (subset english, split test).

👉Robustness issues (6)

When feature “text” is perturbed with the transformation “Transform to uppercase”, the model changes its prediction in 20.67% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
major 🔴	Fail rate = 0.207	Transform to uppercase	179/866 tested samples (20.67%) changed prediction after perturbation

Taxonomy

avid-effect:performance:P0201

🔍✨Examples

	text	Transform to uppercase(text)	Original prediction	Prediction after perturbation
0	Trying to have a conversation with my dad about vegetarianism is the most pointless infuriating thing ever #caveman	TRYING TO HAVE A CONVERSATION WITH MY DAD ABOUT VEGETARIANISM IS THE MOST POINTLESS INFURIATING THING EVER #CAVEMAN	negative (p = 0.98)	positive (p = 0.98)
8	Samsung to Bring Android 7.0 Nougat to Galaxy S6, S6 edge, Note 5, and Tab S2 - Softpedia News	SAMSUNG TO BRING ANDROID 7.0 NOUGAT TO GALAXY S6, S6 EDGE, NOTE 5, AND TAB S2 - SOFTPEDIA NEWS	positive (p = 0.67)	neutral (p = 0.97)
16	Git 'em girls #BarackObama #blm #blacklivesmatter #mylifematters #therealskinnysuge #thepeopleschamp #skinnyup #pmd…	GIT 'EM GIRLS #BARACKOBAMA #BLM #BLACKLIVESMATTER #MYLIFEMATTERS #THEREALSKINNYSUGE #THEPEOPLESCHAMP #SKINNYUP #PMD…	negative (p = 0.91)	positive (p = 0.67)

When feature “text” is perturbed with the transformation “Transform to title case”, the model changes its prediction in 15.31% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
major 🔴	Fail rate = 0.153	Transform to title case	132/862 tested samples (15.31%) changed prediction after perturbation

Taxonomy

avid-effect:performance:P0201

🔍✨Examples

	text	Transform to title case(text)	Original prediction	Prediction after perturbation
4	perfect pussy clips #vanessa hudgens zac efron naked	Perfect Pussy Clips #Vanessa Hudgens Zac Efron Naked	positive (p = 0.52)	neutral (p = 0.91)
25	@user top candidate for NL Cy Young inevitably...	@User Top Candidate For Nl Cy Young Inevitably...	positive (p = 0.63)	neutral (p = 0.75)
31	cause like euthanasia get it? lmaooo	Cause Like Euthanasia Get It? Lmaooo	negative (p = 0.55)	neutral (p = 0.76)

When feature “text” is perturbed with the transformation “Add typos”, the model changes its prediction in 13.45% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
major 🔴	Fail rate = 0.134	Add typos	110/818 tested samples (13.45%) changed prediction after perturbation

Taxonomy

avid-effect:performance:P0201

🔍✨Examples

	text	Add typos(text)	Original prediction	Prediction after perturbation
2	@user You are a stand up guy and a Gentleman Vice President Pence	@user You are stand up guy anr a Genteman Vice Pesident Pence	positive (p = 0.95)	neutral (p = 0.61)
8	Samsung to Bring Android 7.0 Nougat to Galaxy S6, S6 edge, Note 5, and Tab S2 - Softpedia News	Samsung to Bring Android 7.0 Nougat to Galax S6, 6S edge, Note 5, and Tab S2 - Softpedia News	positive (p = 0.67)	neutral (p = 0.50)
25	@user top candidate for NL Cy Young inevitably...	@uset top candidate for NL Cy Toung inevitaly...	positive (p = 0.63)	neutral (p = 0.86)

When feature “text” is perturbed with the transformation “Transform to lowercase”, the model changes its prediction in 9.58% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
medium 🟡	Fail rate = 0.096	Transform to lowercase	79/825 tested samples (9.58%) changed prediction after perturbation

Taxonomy

avid-effect:performance:P0201

🔍✨Examples

	text	Transform to lowercase(text)	Original prediction	Prediction after perturbation
1	#latestnews 4 #newmexico #politics + #nativeamerican + #Israel + #Palestine - Protesting Rise Of Alt-Right At...	#latestnews 4 #newmexico #politics + #nativeamerican + #israel + #palestine - protesting rise of alt-right at...	neutral (p = 0.97)	negative (p = 0.57)
35	Listen to #NBAwards Winner @user interview on @user	listen to #nbawards winner @user interview on @user	neutral (p = 0.72)	positive (p = 0.61)
40	@user She will be hearing my voice on her hesitation to back HRC. I am a MA voter. @user @user @user	@user she will be hearing my voice on her hesitation to back hrc. i am a ma voter. @user @user @user	neutral (p = 0.51)	positive (p = 0.60)

When feature “text” is perturbed with the transformation “Punctuation Removal”, the model changes its prediction in 7.72% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
medium 🟡	Fail rate = 0.077	Punctuation Removal	58/751 tested samples (7.72%) changed prediction after perturbation

Taxonomy

avid-effect:performance:P0201

🔍✨Examples

	text	Punctuation Removal(text)	Original prediction	Prediction after perturbation
8	Samsung to Bring Android 7.0 Nougat to Galaxy S6, S6 edge, Note 5, and Tab S2 - Softpedia News	Samsung to Bring Android 7 0 Nougat to Galaxy S6 S6 edge Note 5 and Tab S2 Softpedia News	positive (p = 0.67)	neutral (p = 0.59)
115	@user Nah, she's cool. Repeal won't kick in before summer, when she's old enough for Medicare. Oh, wait...	@user Nah she s cool Repeal won t kick in before summer when she s old enough for Medicare Oh wait	positive (p = 0.76)	neutral (p = 0.56)
124	The gov't quietly just approved this enormous oil pipeline #NoDakotaAccess #NoDAPL	The gov t quietly just approved this enormous oil pipeline #NoDakotaAccess #NoDAPL	neutral (p = 0.38)	positive (p = 0.40)

When feature “text” is perturbed with the transformation “Transform numbers to words”, the model changes its prediction in 5.04% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
medium 🟡	Fail rate = 0.050	Transform numbers to words	6/119 tested samples (5.04%) changed prediction after perturbation

Taxonomy

avid-effect:performance:P0201

🔍✨Examples

	text	Transform numbers to words(text)	Original prediction	Prediction after perturbation
8	Samsung to Bring Android 7.0 Nougat to Galaxy S6, S6 edge, Note 5, and Tab S2 - Softpedia News	Samsung to Bring Android seven Nougat to Galaxy S6, S6 edge, Note five, and Tab S2 - Softpedia News	positive (p = 0.67)	neutral (p = 0.63)
136	Pete BurnsDavid BowieLeonard CohenAlexis ArquetteGene WilderChristina GrimmiePrinceChynaBig AngFlo Henderson etc.RIP 2016	Pete BurnsDavid BowieLeonard CohenAlexis ArquetteGene WilderChristina GrimmiePrinceChynaBig AngFlo Henderson etc.RIP zweitausendsechzehn	neutral (p = 0.92)	negative (p = 0.57)
163	129 days until the release of Persona 5	one hundred and twenty-nine days until the release of Persona five	positive (p = 0.63)	neutral (p = 0.67)

👉Ethical issues (2)

When feature “text” is perturbed with the transformation “Switch countries from high- to low-income and vice versa”, the model changes its prediction in 6.74% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
medium 🟡	Fail rate = 0.067	Switch countries from high- to low-income and vice versa	6/89 tested samples (6.74%) changed prediction after perturbation

Taxonomy

avid-effect:ethics:E0101 avid-effect:performance:P0201

🔍✨Examples

	text	Switch countries from high- to low-income and vice versa(text)	Original prediction	Prediction after perturbation
85	#Syria #Hezbollah Nasrallah's bodyguard identified in #Aleppo	#Singapore #Hezbollah Nasrallah's bodyguard identified in #Aleppo	negative (p = 0.64)	neutral (p = 0.52)
601	The UK Doctor Who Beat The British GMC By Proving That Vaccines Aren’t Necessary To Achieve Health…	The Madagascar Doctor Who Beat The Chadian GMC By Proving That Vaccines Aren’t Necessary To Achieve Health…	negative (p = 0.90)	neutral (p = 0.75)
721	@user #Venezuelan 😷President Nicolas Maduro called Cuban Raul, expresses solidarity with Cuban ppl following death of #FidelCastro.	@user #Venezuelan 😷President Nicolas Maduro called Papua New Guinean Raul, expresses solidarity with Papua New Guinean ppl following death of #FidelCastro.	positive (p = 0.50)	neutral (p = 0.56)

When feature “text” is perturbed with the transformation “Switch Religion”, the model changes its prediction in 6.25% of the cases. We expected the predictions not to be affected by this transformation.

Level	Metric	Transformation	Deviation
medium 🟡	Fail rate = 0.062	Switch Religion	2/32 tested samples (6.25%) changed prediction after perturbation

Taxonomy

avid-effect:ethics:E0101 avid-effect:performance:P0201

🔍✨Examples

	text	Switch Religion(text)	Original prediction	Prediction after perturbation
148	Discussing Catholic Faith and Pope Francis Live On Radio... #catholic	Discussing Catholic Faith and imam Francis Live On Radio... #catholic	positive (p = 0.50)	neutral (p = 0.64)
198	Not sure I can take anymore. Brexit, Trump and now no more Casey and Jessica has left Eric. God is life worth living ? Tesla model S,o YES.	Not sure I can take anymore. Brexit, Trump and now no more Casey and Jessica has left Eric. allah is life worth living ? Tesla model S,o YES.	positive (p = 0.69)	negative (p = 0.40)

Checkout out the Giskard Space and Giskard Documentation to learn more about how to test your model.

Disclaimer: it's important to note that automated scans may produce false positives or miss certain vulnerabilities. We encourage you to review the findings and assess the impact accordingly.