=Paper=
{{Paper
|id=Vol-2982/paper-5
|storemode=property
|title=Are Knowledge Graph Embedding Models Biased, or Is it the Data That They Are Trained on?
|pdfUrl=https://ceur-ws.org/Vol-2982/paper-5.pdf
|volume=Vol-2982
|authors=Wessel Radstok,Melisachew Wudage Chekol,Mirko T. Schäfer
|dblpUrl=https://dblp.org/rec/conf/semweb/RadstokCS21
}}
==Are Knowledge Graph Embedding Models Biased, or Is it the Data That They Are Trained on?==
https://ceur-ws.org/Vol-2982/paper-5.pdf
Are knowledge graph embedding models biased,
or is it the data that they are trained on?
Wessel Radstok1 , Melisachew Wudage Chekol1 , and Mirko Tobias Schäfer2
1
Data Intensive Systems Group, Utrecht University
2
Department of Media and Culture Studies, Utrecht University
Abstract. Recent studies on bias analysis of knowledge graph (KG) em-
bedding models focus primarily on altering the models such that sensitive
features are dealt with differently from other features. The underlying
implication is that the models cause bias, or that it is their task to solve
it. In this paper we argue that the problem is not caused by the mod-
els but by the data, and that it is the responsibility of the expert to
ensure that the data is representative for the intended goal. To support
this claim, we experiment with two different knowledge graphs and show
that the bias is not only present in the models, but also in the data.
Next, we show that by adding new samples to balance the distribution
of facts with regards to specific sensitive features, we can reduce the bias
in the models.3
1 Introduction
For several days in early July 2018, Google and Apple’s search assistants wrong-
fully reported that the man behind the Marvel comic books, Stan Lee, had passed
away .4 It did not take long for news articles to start popping up noting the un-
justified death declaration. Although Google and Apple never officially reported
on this issue, its source is likely traced back to Wikidata. On June 27th, a Wiki-
data user ran their own script made to parse data from Wikipedia and insert
them as claims into Wikidata. This script then mistakenly pronounced Stan Lee
dead. Other users soon corrected the error, which resulted in an edit war that
became so bad that the page had to be temporarily locked against vandalism.
This is not the only occurrence of incorrect information in knowledge graphs
causing issues in downstream search queries. In the second half of 2018, the for-
mer Guantanamo Bay detainee Omar Khadr was incorrectly reported by Google
search for the query ’Canadian Soldiers’. Again the cause was the script written
by the aforementioned user. Although the issue was quickly resolved after online
outrage, it cropped up twice more over a period of several months. It eventually
led Google to take manual action to fix the knowledge graph.5
3
Copyright © 2021 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0)
4
https://www.cinemablend.com/news/2444550/
siri-is-telling-people-stan-lee-died-yesterday
5
https://www.cbc.ca/news/technology/
omar-khadr-google-search-knowledge-graph-scheer-russia-1.4999775
2 M. Chekol, W. Radstok and M. Schäfer
In addition to the presence of incorrect information in knowledge graphs due
to either an error in the KG construction or intentionally supplied by content
curators, KGs can also be incomplete. As an example, in Freebase [2], over 70%
of person entities have no known place of birth and over 75% have no known
nationality [8]. In Wikidata [14], we observe a similar behavior, for instance,
over 97% of humans have no known religion and over 83% of humans have
no known spoken, written or signed languages. Subsets of both Wikidata and
Freebase have been widely used for testing knowledge graph completion mod-
els. However, these subsets do not take into account the incompleteness of the
KGs and are prepared in a way to test solely the accuracy of models. However,
if the subsets are incomplete (or unbalanced), the models can be biased. For
instance, the Wikidata12K [11, 6] dataset contains 80% male and 20% female
politicians. Clearly, this dataset is unbalanced and a model trained on it will
likely overrepresent men in its predictions.
Indeed this is shown in our experiments using the TransE [3] model; when
asked to predict people most likely to be politicians, the top 100 ranked answers
contain just 12.4% while the remaining 83.6% are male. In order to mitigate
such biased predictions, recently there has been a growing effort towards adapt-
ing/extending KG completion models [4, 9, 10]. These studies on bias analysis of
KG embedding models focus primarily on altering the models such that sensitive
features (such as gender, sexual orientation, etc) are dealt with differently from
other features. The underlying implication is that the models cause bias, or that
it is their task to solve it. However, we found out that the datasets on which
the models are trained on are biased/unbalanced. Although algorithms for the
automatic balancing of data do exist [5], these are not trivial to apply to graph
datasets. Our experiments showed unsatisfactory results using these methods.
Furthermore, adapting models to remove bias means that the resulting em-
beddings will be bias-neutral with regards to the strength of the model used.
That is, removing bias requires a bias detection model and the extent of the the
bias removed depends on how much bias is detected. As a result, embeddings
are not truly neutral: a more powerful model might still be able to detect biases.
Therefore we argue that a domain expert must remain in the loop.
In this work, we address the problem by working directly on the data rather
than altering KG embedding models. In other words, we investigate a new ap-
proach in order to balance (mitigate bias) a given dataset: we automatically
extend a dataset by extracting additional facts to complete missing values of
sensitive features. Moreover, so as to motivate the proposed approach, we car-
ried out a comprehensive analysis of the distribution of sensitive features in
Wikidata highlighting various skewed data distributions.
2 Related Work
We group the related work into two classes of bias analysis: (i) knowledge graphs
and (ii) embedding models.
Title Suppressed Due to Excessive Length 3
Bias analysis of knowledge graphs. [7] proposes methods to trace the provenance
of crowdsourced fact checking to enable bias transparency rather than aiming
at eliminating bias from a KG. Furthermore, they investigate how paid crowd-
sourcing can be used to understand contributors’ implicit bias. Specifically, they
recruit click workers to verify controversial facts and study them as they do so.
I.e., they track what search engines are used and which position the URL used
to validate was ranked in the result page. An example verification task is the
question of whether Catalonia is a part of Spain or an independent country.
The paper proposes adding both facts to the knowledge graph, with a statement
testifying how much support there is for each fact.
[15] introduces ProWD, a framework and tool for profiling the completeness
of Wikidata. Completeness measure is based on Class-Facet-Attribute (CFA)
profiles. For example one could compare how often the attribute ”educated at”
or”date of birth” compare between male, German computer scientists, and fe-
male, Indonesian computer scientists.
Bias analysis of embedding models. Bourli et al. [4] present an analysis method
for investigating gender bias with regards to occupation in entity embeddings.
Specifically, they subtract the male embedding from the female entity embed-
ding to get the bias vector. Projecting an occupation on this vector then gives
them the bias in this occupation. Furthermore, they introduce a de-biasing ap-
proach that generates new de-biased embedding vectors from the existing one
by subtracting it from the bias vector.
[10] conduct experiments on Wikidata and Freebase, and show that harmful
social biases related to professions are encoded in the embeddings with respect
to gender, religion, ethnicity and nationality. They first explain how traditional
word embeddings metrics do not apply to KG embeddings due to the transfor-
mations applied. They then provide a method for evaluating bias. Their method
operates through increasing/decreasing an entities score of a sensitive attribute
(e.g., make it more male and less female) and then recording how the likelihood
of a certain target triple being true changes (e.g., whether they are a nurse of a
lawyer). As a followup, the authors present a novel approach to KG embedding
where embeddings are trained to be neutral with respect to sensitive features
using an adversarial loss function [9] . To achieve this, they add a neural-network
based classifier to the scoring function: scores are penalized when this classifier
can predict the value of the sensitive attribute from the existing embedding.
However, this means that the embeddings are only neutral with respect to the
power of the model: a more powerful model might be able to infer the sensitive
values.
These (and other initiatives) indicate that there is a growing attention to
bias in knowledge graphs, and efforts to make bias visible. As knowledge graphs
often are collaborative repositories, it is relevant to provide users with accessible
means for identifying possible bias. The examples above are helpful but limited
in two ways: they either are valid for a specific knowledge graph, and/or a lim-
ited number of attributes. A general framework might provide more possibilities
to map bias in knowledge graphs and enable users to become aware of the dis-
4 M. Chekol, W. Radstok and M. Schäfer
Fig. 1: How many humans have at least one occurrence of a given property?
tribution of items and attributes in a given knowledge graph. With their own
subject specific expertise, these users can then decide which bias is problematic,
and how to address it.
3 Wikidata Completeness Analysis
Wikidata is the large, open knowledge graph which acts as central storage for the
structured data of other Wikimedia projects such as Wikipedia. Data is stored
as claims or triples, containing a subject item, a property and a value. Values
are entities or literals such as a quantity, a string or even a coordinate. Items
are identified through URIs starting with ’Q’ (e.g. Q22686 for Donald Trump)
and properties are identified through URIs starting with ’P’ (e.g., P40 for Child ).
Claims can be contextualized with additional data such as sources (for the data),
ranks (in case of multiple values for a property) and qualifiers (e.g., to note that
a fact was true at a specific point in time, or that a fact is disputed). A claim
and its additional data are collectively referred to as a statement.
3.1 Completeness
We investigated how several properties were distributed among the class of hu-
mans in Wikidata. An item x is a human when it is an instance of (P31) human
(Q5), i.e., item x must have the claim (x, P31, Q5). Using the Wikidata dump
from 2021/03/31, we extracted 9,028,271 such items. We will now give a brief
overview of some of our preliminary findings.
To begin we, for each item in the subset we counted whether or not a property
occurs among its claims. This gives us an overview of how often a property
occurs at least once. The result is displayed in Figure 1. Ignoring the predicate
instance of, which per our definition is present on all humans, the most occurring
predicates are sex or gender (P21), occupation (P106), and given name (P735).
Title Suppressed Due to Excessive Length 5
Fig. 2: How many humans have a label specified in the given language.
These occur on 7,079,543 (78%), 6,359,256 (70%) and 5,635,238 (62%) humans
respectively.
Additionally, we counted the number of languages each item had a label in.
This gives us an overview of how complete Wikidata is over several languages.
The result of this is displayed in Figure 2. Expectedly, the most common lan-
guage is English, with 8.517.283 (94%) humans having an English label. More
unexpectedly however is the fact that, in spite of being a small country with
only 17 million inhabitants, the second most common label language is Dutch
with 7.785.518 (86%) humans having a Dutch label.
Next, we can look at the distribution of object entities for a given predicate.
I.e., given a predicate such as place of death (P20) we can count how many
people have object values such has Moscow or Paris. From this data, we have
created a bar graph for a selection of predicates in Figure 3.
Looking at this data, it is immediately clear that it is not representative
of the common population. For instance, the most common occupation by far
is researcher (20%). Yet in reality, even in the USA only around 2% of the
population has a PhD.6 We of course understand that an encyclopedia covers
persons of interest and not the general population. Hence it is logical that there
is a bias. However, the problematic bias is not the overrepresentation of scholars
but the overrepresentation of of white male scholars at western universities. If
we want to inquire to what extend the population of researchers in Wikipedia is
skewed we need to inquire about the presence of other occupations for persons
of interest for an encyclopedia, such as athletes, activists, politicians, engineers
and inventors.
We hypothesize that there are two main sources of bias present in this data.
The first is availability bias, i.e., much of the data present in Wikidata is there
because it could be easily imported. For instance through the use of bots. The
second is interest bias, where the interests of the people who work on Wikidata
end up deciding what content will dominate the dataset. Examples of this bias are
the most common occupation being researcher (imported through article papers)
and the second most common place of death being a concentration camp.
6
https://data.worldbank.org/indicator/SE.TER.CUAT.DO.ZS?locations=US
6 M. Chekol, W. Radstok and M. Schäfer
(a) Most common place of birth (left) and place of death (right) entities.
(b) Most common ethnic groups (left) and languages spoken (right) entities.
(c) Most common occupations (left) and religions (right) entities.
Fig. 3: Overview of the most common values for several predicates in Wikidata.
3.2 Spatiotemporal Analysis
Temporal information in Wikidata present itself in two ways. Firstly, predicates
can directly have timestamps as their object value. For instance, the date of
birth of a person. All predicates that can have a timestamp as object value
must be instances of (P31) Wikidata property with datatype ’time’ (Q18636219).
There are 34 such predicates. Secondly, temporal information can be included
in any other predicate through the use of qualifiers. I.e., predicates start time
(P569) and end time (P570) can be applied to a triple through reification to add
temporal information to that triple.
Since we are interested in how humans are represented in Wikidata, we re-
strict spatiotemporal analysis to the human class. Specifically, we ground data
in space by looking at a persons place of birth (P19 ) and in time by looking
at the place of birth (P569 ). Through this we can analyse the completeness of
Title Suppressed Due to Excessive Length 7
(a) Number of countries with at least one fact in the given century.
(b) Comparison between the most common ethnic groups listed in Wikidata in the
18th (left) and 21st (right) century.
Fig. 4: Overview of some select metrics of Wikidata temporal metrics.
Wikidata over time. Some results are displayed in Figure 4. We observe that the
further we go back in time, the less distinct countries are observed in Wikidata.
I.e., facts seem to be more based on a few countries. Addtionally, we investigate
the occurrences of the most common ethnic groups listed in Wikidata. Interest-
ingly, the use of ethnic group seems to have fallen out of favour for people born
more recently. In the 18th century the most common ethnic groups was Greeks
with over 400 occurrences, whereas in the 21st century the most common ethnic
group is African American with just over 100 occurrences.
4 Bias Analysis of Knowledge Graph Embedding Models
In this section we perform bias analysis of the knowledge graph embedding mod-
els. Specifically, we analyze the effect of balancing the data on link prediction
performance. For this task we utilize two popular models, TransE [3] and Dist-
Mult [16]. We perform our experiments on two state-of-the-art knowledge graphs.
The first is Wikidata12k, a subset of Wikidata extracted by [11, 6]. The second
is DBP15k, a subset of DBpedia [1] originally created by [13] to test Entity-
Alignment models. As we are interested in link prediction rather than entity
alignment, we select a single instance of the dataset (the English version) and
perform our experiments on it. All of our code is available on Github 7
7
https://github.com/wradstok/KGE-bias-analyzer
8 M. Chekol, W. Radstok and M. Schäfer
4.1 Embedding Models
For a triple (s, p, o), let (es , ep , eo ) denote its embedding vectors. Taking a KG
and a random initialization of the vectors as an input, a vector representation
of the KG is gradually learned using a scoring function φ(s, p, o). The scoring
function should reflect how well the embedding captures the semantics of the KG.
The learned embeddings can be used in tasks such as classification, clustering,
and link prediction. In this work, we are focussed on the last. Link prediction is
the task of predicting the most likely element given a tuple where one element
is missing, e.g., given a triple (s,p,? ), to predict the most likely object entity.
The most popular embedding model is TransE (Translating embeddings).
Its scoring function is based on the intuition that the the subject and object
vector should be close together after adding the predicate vector. Its scoring
function is written as φ(s, p, o) = ||es + ep − eo ||1,2 . While being very powerful,
it has limited expressiveness due to its is simplicty. Therefore, we also perform
experiments with DistMult, a multiplicative model. Its scoring function is written
as φ(s, p, o) = ||es ∗ ep ∗ eo ||1,2, . In our experiments we do not use pre-trained
models and instead train the embeddings from scratch.
4.2 SMOTE
One way of balancing datasets is to use Synthetic Minority Over-sampling TEch-
nique (SMOTE) [5]. SMOTE is an over-sampling technique that allows one to
construct new examples of a given class based on existing examples in order to
address imbalances in the dataset. An example would be oversampling female
football players. However, SMOTE is not intended for graph datasets and as
such is not trivial to apply to knowledge graphs while maintaining the underly-
ing structure.
One way to apply SMOTE to graph data is through embedding the graph
first. We use this approach to evaluate how well SMOTE is suited for our sce-
nario. Our method is as follows. After obtaining the embeddings, we create a
categorical variable with a category for each possible combination of sensitive
features. In the case of 5 occupations and 2 genders, this implies 10 categories.
The embeddings vectors are then combined with this categorical value associated
with the sample it represents. Finally, we instruct SMOTE to generate the max-
imum number of samples for each possible entry using the python imbalanced-
learn library [12], i.e., given that there are 1850 male association football players,
we create both male and female physicists until there are 1850 of each.
However, preliminary experiments found that this method did not suffice
for generating a balanced set of embeddings. Applying our evaluation method
to datasets produced by above procedure did not create balanced predictions.
We hypothesize that it is because SMOTE generates new examples based on
existing biases. By interpolating new ‘female’ examples from existing female
embeddings, we are only creating new examples in the same cluster. That means
that locations in the embedding space which are already female become much
more so. Therefore, we instead extend the datasets by sampling additional triples
from the original knowledge graphs.
Title Suppressed Due to Excessive Length 9
Dataset # Triples # Entities # Pred. # Men # Women
Wikidata12k (original) 38,970 12,848 25 4,905 717
Wikidata12k (balanced) 51,682 15,957 25 4,905 3,610
DBpedia15k (original) 92,746 18,716 206 6,767 1,087
DBpedia15k (balanced) 95,827 27,459 206 5,916 5,917
Table 1: Dataset statistics
4.3 Sampling process
We enrich the original knowledge graphs by adding female triples, i.e. extra
triples with female entities as subject. The data is enriched in such a way that
the number of men and women associated with each of the top 5 most common
occupations becomes approximately equal. The triples are obtained from the
complete Wikidata and DBpedia datasets.
To ensure that the new triples have healthy connectivity with regards to the
rest of the graph this is done in a three step process. Firstly, all women with the
required occupations are selected from the complete knowledge graph. Secondly,
from this selection the women which have the largest number of predicates which
are also in the original dataset are picked. Finally, we select the women whose
object values are already in the graph. The last step ensures that we do not add
object entities which occur only a few times, and only with women.
4.4 Wikidata12k
Wikidata12k does not contain any information about gender or occupation. How-
ever, we can look up this data by querying the original Wikidata knowledge
graph. As Wikidata12k is originally a temporal knowledge graph, we strip out
the temporal information and remove and duplicate triples that may be created
by this process.
The five most common occupations are association football player Q937857
(1867), politician Q82955 (918), actor Q33999 (211), writer Q36180 (184) and
physicist Q169470 (143). These occupations are not uniformly distributed with
regards to gender: there are only a handful of women football players, and there
is not a single woman physicist in the entire dataset.
In total, we add around 10,000 triples with female entities as subject to the
Wikidata12k knowledge graph, resulting a new graph over 50,0000 triples. This
increases the average number of mentions as subject (i.e., the average number of
outlinks) per female entity from 3.49 in the original graph to 4.74 in the balanced
graph. However, the number of outlinks still falls short of that of men, which is
5.41.
4.5 DBP15k
DBP15k is a subset of DBpedia [1] created by [13] to test Entity-Alignment
models. The majority of predicates in DBP15k have very few triples associated
10 M. Chekol, W. Radstok and M. Schäfer
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Officeholder 1803 180 9.1% 85 15 15.0% 6.9
Athlete 1142 6 0.5% 100 0 0.0% -0.5
Royalty 569 235 29.2% 60 40 40.0% 10.8
Sportsmanager 225 0 0.0% 100 0 0.0% 0.0
Scientist 216 6 2.7% 95 5 5.0% 2.3
Total 3955 427 9.7% 440 60 12.0% -
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Officeholder 1498 1770 54.2% 16 83 83.8% 29.7
Athlete 990 1320 57.1% 55 45 45.0% -12.1
Royalty 472 596 55.8% 25 74 74.7% 18.9
Sportsmanager 188 31 14.2% 85 15 15.0% 0.8
Scientist 195 225 53.6% 32 67 67.7% 14.1
Total 3343 3942 54.1% 213 284 57.1% -
Table 2: Comparison between the male/female distribution and the resulting
TransE model predictions in the original DBPedia15k dataset (top) and our
balanced dataset (bottom). Difference column contains the difference (in per-
centage points) between the % of women in the predicted and the % of women
predicted.
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Politician 619 96 13.4% 80 20 20.0% 6.6
Writer 100 33 24.8% 74 25 25.3% 0.5
Actor 67 89 57.1% 38 61 61.6% 4.6
Football player 1465 14 0.9% 98 2 2.0% 1.1
Physicist 108 2 1.8% 98 2 2.0% 0.2
Total 2359 234 9.0% 388 110 22.1% -
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Politician 644 331 33.9% 53 47 47.0% 13.1
Writer 111 59 34.7% 25 74 74.7% 40.0
Actor 71 93 56.7% 33 66 66.7% 10.0
Football player 1455 1427 49.5% 8 91 91.9% 42.4
Physicist 122 44 26.5% 56 44 44.0% 17.5
Total 2403 1954 44.8% 175 322 64.8% -
Table 3: Comparison between the male/female distribution and the resulting
TransE model predictions in the original Wikidata12k dataset (top) and our
balanced dataset (bottom). Difference column contains the difference (in per-
centage points) between the % of women in the predicted and the % of women
predicted.
Title Suppressed Due to Excessive Length 11
with them. To prevent the graph from being too sparse for an embedding model
to learn we delete all predicates which occur less than 50 times.
DBpedia does not store any information about peoples sex or gender in a
structured way. I.e., although a person can be of rdf:type of Man or Woman,
manual inspection of the data did not reveal that this information was consis-
tently present. However, most entities do contain their Wikidata identifiers. Since
Wikidata does list peoples gender, we determine a persons gender by querying
Wikidata for the given identifiers.
The five most common occupations are OfficeHolder (2508), Athlete (1436),
Royalty (1002), SportsManager (288), and Scientist (282). Like Wikidata12k, the
male/female ratio in these occupations is unbalanced, skewing heavily towards
men. In addition to balancing the data by adding additional samples, we remove
some male entities and their triples to create the balanced dataset.
4.6 Evaluation
To evaluate whether an embedding model contains bias with regards to gen-
der and occupation we perform the following procedure. Firstly, we count the
fraction of men and women that have a certain occupation (P106 ) x. Then, we
ask the model to predict the n most likely entities for the query (?, P106, x).
If the fraction of men or women returned is consistently larger than that the
fraction present in the data, the model is biased. Specifically, when more men
are predicted the model is biased against women and vice versa. If this bias is
only present in the unbalanced dataset and not the balanced datasets, then the
model reflects the data it has been trained on. However, if the bias is present
in both scenarios, the models are either inherently biased or manage to pick up
some form of bias in the data which is not reflected in our analysis.
4.7 Results
Our results are displayed in Tables 2 and 3 for DBpedia and Wikidata12k respec-
tively using TransE [3], and in Tables 4 and 5 using DistMult [16]. We observe
that in both original datasets, the percentage of women predicted is very low
and close to the percentage of women in the dataset. The largest difference is ob-
served on the occupation Royalty in the DBpedia15k dataset. Here the difference
is just over 10 percentage points.
When we extend our view to the balanced datasets, we find that the percent-
age of women predicted has moved upwards with the percentage of women in the
dataset. Balancing the datasets thus helps with improving the representation of
minority classes in the model output. However, we do observe that the absolute
differences between the number of men in the dataset, and the number of men
predicted (and for women) has increased, suggesting that the model has become
less accurate.
Even so, we believe a more likely explanation to be that the larger number
of entities predicted induces more variance in the predictions. This explanation
is strengthened by the fact that the difference is smaller when using DistMult,
12 M. Chekol, W. Radstok and M. Schäfer
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Officeholder 1803 180 9.1% 87 13 13.0% 3.9
Athlete 1142 6 0.5% 100 0 0.0% 0.5
Royalty 569 235 29.2% 68 32 32.0% 2.8
Sportsmanager 225 0 0.0% 100 0 0.0% 0.0
Scientist 216 6 2.7% 96 4 4.0% 1.3
Total 3955 427 9.7% 451 49 9.8% -
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Officeholder 1498 1770 54.2% 47 53 53.0% -1.2
Athlete 990 1320 57.1% 78 22 22.0% -35.14
Royalty 472 596 55.8% 39 61 61.0% 5.2
Sportsmanager 188 31 14.2% 88 12 12.0% -2.2
Scientist 195 225 53.6% 51 49 49.0% -4.6
Total 3343 3942 54.1% 303 197 39.4% -
Table 4: Comparison between the male/female distribution and the resulting
DistMult model predictions in the original DBpedia15k dataset (top) and
our balanced dataset (bottom). Difference column contains the difference (in
percentage points) between the % of women in the predicted and the % of
women predicted.
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Politician 619 96 13.4% 83 17 17.0% 3.6
Writer 100 33 24.8% 72 28 28.0% 3.2
Actor 67 89 57.1% 50 50 50.0% -7.1
Football player 1465 14 0.9% 99 1 1.0% 0.1
Physicist 108 2 1.8% 95 5 5.0% 3.2
Total 2359 234 9.0% 399 101 20.2%
Data Prediction
Men Women Women (%) Men Women Women (%) Diff (p.p.)
Politician 644 331 33.9% 51 49 49.0% 15.1
Writer 111 59 34.7% 39 61 61.0% 26.3
Actor 71 93 56.7% 37 63 63.0% 6.3
Football player 1455 1427 49.5% 50 50 50.0% 0.5
Physicist 122 44 26.5% 42 58 58.0% 31.5
Total 2403 1954 44.8% 219 281 56.2%
Table 5: Comparison between the male/female distribution and the resulting
DistMult model predictions in the original Wikidata12k dataset (top) and
our balanced dataset (bottom). Difference column contains the difference (in
percentage points) between the % of women in the predicted and the % of
women predicted.
Title Suppressed Due to Excessive Length 13
which is a more expressive model and can thus model the information more
accurately.
Another point of note is the observation that on both datasets and almost
all occupations, the sign of the difference between the percentage of women pre-
dicted and the percentage of women in the dataset is mostly positive. This means
that women are actually overrepresented in the models predictions, indicating
that the model is actually less biased than the data.
5 Conclusion
In this paper we proposed a new approach to mitigate bias in knowledge graphs
embedding models by leveraging the distribution of the datasets in which the
models are trained on. Specifically, rather than adapting models to mitigate bias,
we instead analyze and augment the data that is fed into the model. We carried
out several experiments using state of the art embedding models (namely, TransE
and DistMult) and two knowledge graphs (namely DBpedia and Wikidata) and
showed that balancing the data with regards to specific sensitive features (e.g.
gender and occupation) improves the overall prediction capabilities of the mod-
els. Additionally, to motivate our work, we have done a completeness analysis of
Wikidata using a number of sensitive features.
As a future work, we will extend the proposed approach to build a system that
takes as an input a dataset and a selection of sensitive features and automatically
balances the data with respect to the given features.
References
1. Bizer, C., Lehmann, J., Kobilarov, G., Auer, S., Becker, C., Cyganiak, R., Hell-
mann, S.: Dbpedia-a crystallization point for the web of data. Journal of web
semantics 7(3), 154–165 (2009)
2. Bollacker, K., Evans, C., Paritosh, P., Sturge, T., Taylor, J.: Freebase: a collabo-
ratively created graph database for structuring human knowledge. In: Proceedings
of the 2008 ACM SIGMOD international conference on Management of data. pp.
1247–1250 (2008)
3. Bordes, A., Usunier, N., Garcia-Duran, A., Weston, J., Yakhnenko, O.: Translating
embeddings for modeling multi-relational data. Advances in neural information
processing systems 26 (2013)
4. Bourli, S., Pitoura, E.: Bias in knowledge graph embeddings. In:
2020 IEEE/ACM International Conference on Advances in So-
cial Networks Analysis and Mining (ASONAM). pp. 6–10 (2020).
https://doi.org/10.1109/ASONAM49781.2020.9381459
5. Chawla, N.V., Bowyer, K.W., Hall, L.O., Kegelmeyer, W.P.: Smote: synthetic mi-
nority over-sampling technique. Journal of artificial intelligence research 16, 321–
3x57 (2002)
6. Dasgupta, S.S., Ray, S.N., Talukdar, P.: Hyte: Hyperplane-based temporally aware
knowledge graph embedding. In: Proceedings of the 2018 conference on empirical
methods in natural language processing. pp. 2001–2011 (2018)
14 M. Chekol, W. Radstok and M. Schäfer
7. Demartini, G.: Implicit bias in crowdsourced knowledge graphs. In: Com-
panion Proceedings of The 2019 World Wide Web Conference. p. 624–630.
WWW ’19, Association for Computing Machinery, New York, NY, USA (2019).
https://doi.org/10.1145/3308560.3317307, https://doi.org/10.1145/3308560.
3317307
8. Dong, X., Gabrilovich, E., Heitz, G., Horn, W., Lao, N., Murphy, K., Strohmann,
T., Sun, S., Zhang, W.: Knowledge vault: A web-scale approach to probabilistic
knowledge fusion. In: Proceedings of the 20th ACM SIGKDD international con-
ference on Knowledge discovery and data mining. pp. 601–610 (2014)
9. Fisher, J., Mittal, A., Palfrey, D., Christodoulopoulos, C.: Debiasing knowledge
graph embeddings. In: Proceedings of the 2020 Conference on Empirical Methods
in Natural Language Processing (EMNLP). pp. 7332–7345. Association for Compu-
tational Linguistics, Online (Nov 2020). https://doi.org/10.18653/v1/2020.emnlp-
main.595, https://www.aclweb.org/anthology/2020.emnlp-main.595
10. Fisher, J., Palfrey, D., Christodoulopoulos, C., Mittal, A.: Measuring social bias
in knowledge graph embeddings. arXiv preprint arXiv:1912.02761 todo (2019)
11. Leblay, J., Chekol, M.W.: Deriving validity time in knowledge graph. In: Compan-
ion Proceedings of the The Web Conference 2018. pp. 1771–1776 (2018)
12. Lemaı̂tre, G., Nogueira, F., Aridas, C.K.: Imbalanced-learn: A python toolbox to
tackle the curse of imbalanced datasets in machine learning. Journal of Machine
Learning Research 18(17), 1–5 (2017), http://jmlr.org/papers/v18/16-365
13. Sun, Z., Hu, W., Li, C.: Cross-lingual entity alignment via joint attribute-preserving
embedding. In: d’Amato, C., Fernandez, M., Tamma, V., Lecue, F., Cudré-
Mauroux, P., Sequeda, J., Lange, C., Heflin, J. (eds.) The Semantic Web – ISWC
2017. pp. 628–644. Springer International Publishing, Cham (2017)
14. Vrandečić, D., Krötzsch, M.: Wikidata: a free collaborative knowledgebase. Com-
munications of the ACM 57(10), 78–85 (2014)
15. Wisesa, A., Darari, F., Krisnadhi, A., Nutt, W., Razniewski, S.: Wikidata com-
pleteness profiling using prowd. In: Proceedings of the 10th International Confer-
ence on Knowledge Capture. p. 123–130. K-CAP ’19, Association for Computing
Machinery, New York, NY, USA (2019). https://doi.org/10.1145/3360901.3364425,
https://doi.org/10.1145/3360901.3364425
16. Yang, B., Yih, W.t., He, X., Gao, J., Deng, L.: Embedding entities and relations for
learning and inference in knowledge bases. arXiv preprint arXiv:1412.6575 (2014)