Datasets:

mteb
/

MLQuestions

_id stringlengths 2 6	text stringlengths 3 395	title stringclasses 1 value
C0	In a supervised learning model, the algorithm learns on a labeled dataset, providing an answer key that the algorithm can use to evaluate its accuracy on training data. An unsupervised model, in contrast, provides unlabeled data that the algorithm tries to make sense of by extracting features and patterns on its own.
C1	Set the significance level, , the probability of making a Type I error to be small — 0.01, 0.05, or 0.10. Compare the P-value to . If the P-value is less than (or equal to) , reject the null hypothesis in favor of the alternative hypothesis. If the P-value is greater than , do not reject the null hypothesis.
C2	Minibatch Discrimination is a discriminative technique for generative adversarial networks where we discriminate between whole minibatches of samples rather than between individual samples. This is intended to avoid collapse of the generator.
C3	Eigenvectors can be used to represent a large dimensional matrix. This means that a matrix M and a vector o can be replaced by a scalar n and a vector o. In this instance, o is the eigenvector and n is the eigenvalue and our target is to find o and n.
C4	Machine learning (ML) systems promise disruptive capabilities in multiple industries. Behind the hype, there are three essential risks to analyze when building an ML system: 1) poor problem solution alignment, 2) excessive time or monetary cost, and 3) unexpected behavior once deployed.
C5	There are two types of estimations used: point and interval. A point estimation is a type of estimation that uses a single value, a sample statistic, to infer information about the population. Interval estimation is the range of numbers in which a population parameter lies considering margin of error.
C6	In terms of machine learning, "concept learning" can be defined as: “The problem of searching through a predefined space of potential hypotheses for the hypothesis that best fits the training examples.” — Tom Michell. Much of human learning involves acquiring general concepts from past experiences.
C7	The sample frame, or sample universe, is the data that our sample is drawn from. In the case of the March, 2000 CPS, the sample universe includes all people residing in the US in March, 2000, who were not living in institutional settings. In theory, our sample is drawn from the sample universe.
C8	We reject the null hypothesis when the p-value is less than α. But 0.07 > 0.05 so we fail to reject H0. For example if the p-value = 0.08, then we would fail to reject H0 at the significance level of α=0.05 since 0.08 > 0.05, but we would reject H0 at the significance level of α = 0.10 since 0.08 < 0.10.
C9	Regression is primarily used to build models/equations to predict a key response, Y, from a set of predictor (X) variables. Correlation is primarily used to quickly and concisely summarize the direction and strength of the relationships between a set of 2 or more numeric variables.
C10	For example, an ANOVA test assumes that the variances of different populations are equal (i.e. homogeneous). One example of a test is the Chi-Square Test for Homogeneity. This tests to see if two populations come from the same unknown distribution (if they do, then they are homogeneous).
C11	Least squares also has issues dealing with multicollinearity in data. Ridge regression avoids all of these problems. It works in part because it doesn't require unbiased estimators; While least squares produces unbiased estimates, variances can be so large that they may be wholly inaccurate.
C12	Page 1. Abstract: Structural Vector Autoregressions (SVARs) are a multivariate, linear repre- sentation of a vector of observables on its own lags. SVARs are used by economists to recover economic shocks from observables by imposing a minimum of assumptions compatible with a large class of models.
C13	Time series regression is a statistical method for predicting a future response based on the response history (known as autoregressive dynamics) and the transfer of dynamics from relevant predictors. Time series regression is commonly used for modeling and forecasting of economic, financial, and biological systems.
C14	Control variables are usually variables that you are not particularly interested in, but that are related to the dependent variable. You want to remove their effects from the equation. A control variable enters a regression in the same way as an independent variable - the method is the same.
C15	For example, you could be: 25 years, 10 months, 2 days, 5 hours, 4 seconds, 4 milliseconds, 8 nanoseconds, 99 picosends…and so on. Time is a continuous variable. You could turn age into a discrete variable and then you could count it.
C16	DeepDream is a computer vision program created by Google engineer Alexander Mordvintsev that uses a convolutional neural network to find and enhance patterns in images via algorithmic pareidolia, thus creating a dream-like hallucinogenic appearance in the deliberately over-processed images.
C17	4.3 The method: evolutionary computation. EC is a computational intelligence technique inspired from natural evolution. An EC algorithm starts with creating a population consisting of individuals that represent solutions to the problem. The first population could be created randomly or fed into the algorithm.
C18	Cluster Analysis and Factor Analysis. Latent Class Analysis is similar to cluster analysis. Observed data is analyzed, connections are found, and the data is grouped into clusters. Another difference is that LCA includes discrete latent categorical variables that have a multinomial distribution.
C19	When comparing two groups, you need to decide whether to use a paired test. When comparing three or more groups, the term paired is not apt and the term repeated measures is used instead. Use an unpaired test to compare groups when the individual values are not paired or matched with one another.
C20	In neural networks, Convolutional neural network (ConvNets or CNNs) is one of the main categories to do images recognition, images classifications. Objects detections, recognition faces etc., are some of the areas where CNNs are widely used.
C21	HMMs is the Hidden Markov Models library for Python. It is easy to use, general purpose library, implementing all the important submethods, needed for the training, examining and experimenting with the data models.
C22	These lessons on probability will include the following topics: Samples in probability, Probability of events, Theoretical probability, Experimental probability, Probability problems, Tree diagrams, Mutually exclusive events, Independent events, Dependent events, Factorial, Permutations, Combinations, Probability in
C23	In short, Softmax Loss is actually just a Softmax Activation plus a Cross-Entropy Loss. Softmax is an activation function that outputs the probability for each class and these probabilities will sum up to one. Cross Entropy loss is just the sum of the negative logarithm of the probabilities.
C24	In mathematical optimization and decision theory, a loss function or cost function is a function that maps an event or values of one or more variables onto a real number intuitively representing some "cost" associated with the event. In optimal control, the loss is the penalty for failing to achieve a desired value.
C25	TensorFlow Federated (TFF) is an open-source framework for machine learning and other computations on decentralized data. For example, FL has been used to train prediction models for mobile keyboards without uploading sensitive typing data to servers.
C26	There are two reasons why Mean Squared Error(MSE) is a bad choice for binary classification problems: If we use maximum likelihood estimation(MLE), assuming that the data is from a normal distribution(a wrong assumption, by the way), we get the MSE as a Cost function for optimizing our model.
C27	(e.g. if P=1/256, that's 8 bits.) Entropy is just the average of that information bit length, over all the outcomes. The purpose of log(pi) appearing in Shannon's Entropy is that log(pi) is the only function satisfying the basic set of properties that the entropy function, H(p1,…,pN), is held to embody.
C28	One of the most intuitive explanations of eigenvectors of a covariance matrix is that they are the directions in which the data varies the most. The eigenvectors of the covariance matrix of these data samples are the vectors u and v; u, longer arrow, is the first eigenvector and v, the shorter arrow, is the second.
C29	Joint probability is the probability of two events occurring simultaneously. Marginal probability is the probability of an event irrespective of the outcome of another variable. Conditional probability is the probability of one event occurring in the presence of a second event.
C30	In many situations, the degrees of freedom are equal to the number of observations minus one. Thus, if the sample size were 20, there would be 20 observations; and the degrees of freedom would be 20 minus 1 or 19.
C31	In artificial intelligence, an expert system is a computer system that emulates the decision-making ability of a human expert. Expert systems are designed to solve complex problems by reasoning through bodies of knowledge, represented mainly as if–then rules rather than through conventional procedural code.
C32	In probability theory and statistics, the gamma distribution is a two-parameter family of continuous probability distributions. The exponential distribution, Erlang distribution, and chi-squared distribution are special cases of the gamma distribution.
C33	Selection Sort in CExample of Selection Sort.Algorithm for Selection Sort:Step 1 − Set min to the first location.Step 2 − Search the minimum element in the array.Step 3 – swap the first location with the minimum value in the array.Step 4 – assign the second element as min.Step 5 − Repeat the process until we get a sorted array.More items•
C34	More formally, statistical power is the probability of finding a statistically significant result, given that there really is a difference (or effect) in the population. So, larger sample sizes give more reliable results with greater precision and power, but they also cost more time and money.
C35	The method cannot be considered to derive composite relations. Examples s = ut + 1/2 at2 and 2as = v2 – u2. A formula containing trigonometric function, exponential function, and logarithmic function can not derive from it. The method cannot be used to derive the relationship between more than three quantities.
C36	In probability theory and related fields, a stochastic or random process is a mathematical object usually defined as a family of random variables. Stochastic processes are widely used as mathematical models of systems and phenomena that appear to vary in a random manner.
C37	From Wikipedia, the free encyclopedia. The control variates method is a variance reduction technique used in Monte Carlo methods. It exploits information about the errors in estimates of known quantities to reduce the error of an estimate of an unknown quantity.
C38	Tests for randomness can be used to determine whether a data set has a recognisable pattern, which would indicate that the process that generated it is significantly non-random. These generators do not always generate sequences which are sufficiently random, but instead can produce sequences which contain patterns.
C39	Deep learning neural networks are trained using the stochastic gradient descent optimization algorithm. The learning rate is a hyperparameter that controls how much to change the model in response to the estimated error each time the model weights are updated.
C40	"Degrees of freedom" is commonly abbreviated to df. When this principle of restriction is applied to regression and analysis of variance, the general result is that you lose one degree of freedom for each parameter estimated prior to estimating the (residual) standard deviation.
C41	0:001:38Suggested clip · 98 secondsFind the matrix A given the eigenvalues and eigenvectors - YouTubeYouTubeStart of suggested clipEnd of suggested clip
C42	Long short-term memory (LSTM) is an artificial recurrent neural network (RNN) architecture used in the field of deep learning. LSTM networks are well-suited to classifying, processing and making predictions based on time series data, since there can be lags of unknown duration between important events in a time series.
C43	"Normal" data are data that are drawn (come from) a population that has a normal distribution. This distribution is inarguably the most important and the most frequently used distribution in both the theory and application of statistics. If X is a normal random variable, then the probability distribution of X is.
C44	We use three main types of layers to build ConvNet architectures: Convolutional Layer, Pooling Layer, and Fully-Connected Layer (exactly as seen in regular Neural Networks). We will stack these layers to form a full ConvNet architecture. Example Architecture: Overview.
C45	Advantages of convenience samplingConvenience sampling is vey easy to carry out with few rules governing how the sample should be collected.The relative cost and time required to carry out a convenience sample are small in comparison to probability sampling techniques.More items
C46	In probability, the set of outcomes from an experiment is known as an Event. So say for example you conduct an experiment by tossing a coin. The outcome of this experiment is the coin landing 'heads' or 'tails'. These can be said to be the events connected with the experiment.
C47	To develop or improve your inductive reasoning, focus on the following skills: Paying attention to detail: No one can draw conclusions based on details without first noticing those details; paying attention is crucial to inductive reasoning.
C48	Augmented Analytics This form of analytics is going to play a huge role in analysing data in 2020. Augmented analytics is going to be the future of data analytics because it can scrub raw data for valuable parts for analysis, automating certain parts of the process and making the data preparation process easier.
C49	If your regression model contains independent variables that are statistically significant, a reasonably high R-squared value makes sense. The statistical significance indicates that changes in the independent variables correlate with shifts in the dependent variable.
C50	State–action–reward–state–action (SARSA) is an algorithm for learning a Markov decision process policy, used in the reinforcement learning area of machine learning. The acronym for the quintuple (st, at, rt, st+1, at+1) is SARSA.
C51	How to Detect Omitted Variable Bias and Identify Confounding Variables. You saw one method of detecting omitted variable bias in this post. If you include different combinations of independent variables in the model, and you see the coefficients changing, you're watching omitted variable bias in action!
C52	Typical discriminative models include logistic regression (LR), support vector machines (SVM), conditional random fields (CRFs) (specified over an undirected graph), decision trees, neural networks, and many others.
C53	To find your weighted average, simply multiply each number by its weight factor and then sum the resulting numbers up. For example: The weighted average for your quiz grades, exam, and term paper would be as follows: 82(0.2) + 90(0.35) + 76(0.45) = 16.4 + 31.5 + 34.2 = 82.1.
C54	Moments help in finding AM, standard deviation and variance of the population directly, and they help in knowing the graphic shapes of the population. We can call moments as the constants used in finding the graphic shape, as the graphic shape of the population also help a lot in characterizing a population.
C55	The year is a categorical variable. The ratio between two years is not meaningful which is why its not appropriate to classify it as a quantitative variable.
C56	An embedding is a relatively low-dimensional space into which you can translate high-dimensional vectors. Embeddings make it easier to do machine learning on large inputs like sparse vectors representing words. An embedding can be learned and reused across models.
C57	In Neural network, some inputs are provided to an artificial neuron, and with each input a weight is associated. Weight increases the steepness of activation function. This means weight decide how fast the activation function will trigger whereas bias is used to delay the triggering of the activation function.
C58	The goal of lasso regression is to obtain the subset of predictors that minimizes prediction error for a quantitative response variable. The lasso does this by imposing a constraint on the model parameters that causes regression coefficients for some variables to shrink toward zero.
C59	The log transformation can be used to make highly skewed distributions less skewed. This can be valuable both for making patterns in the data more interpretable and for helping to meet the assumptions of inferential statistics. Figure 1 shows an example of how a log transformation can make patterns more visible.
C60	Probability density function (PDF) is a statistical expression that defines a probability distribution (the likelihood of an outcome) for a discrete random variable (e.g., a stock or ETF) as opposed to a continuous random variable.
C61	Machine learning helps computers understand what they see So computer vision methods nowadays leverage intelligent algorithms and systems. Even to an extent that visual computing has become one of the main fields of this technology's successful application.
C62	The major difference between a traditional Artificial Neural Network (ANN) and CNN is that only the last layer of a CNN is fully connected whereas in ANN, each neuron is connected to every other neurons as shown in Fig. 2.
C63	(mathematics) A symbol representing a product over a set of terms.
C64	While many people use the terms interchangeably, data science and big data analytics are unique fields, with the major difference being the scope. Data science produces broader insights that concentrate on which questions should be asked, while big data analytics emphasizes discovering answers to questions being asked.
C65	Look at normality plots of the data. “Normal Q-Q Plot” provides a graphical way to determine the level of normality. The black line indicates the values your sample should adhere to if the distribution was normal. If the dots fall exactly on the black line, then your data are normal.
C66	When two events are dependent events, one event influences the probability of another event. A dependent event is an event that relies on another event to happen first.
C67	Prior probability shift. Prior probability shift refers to changes in the distribution of the class variable y It also appears with different names in the class variable y. It also appears with different names in the literature and the definitions have slight differences between them.
C68	Values range from 0 to 1, where 0 is perfect disagreement and 1 is perfect agreement. Krippendorff suggests: “[I]t is customary to require α ≥ . 800. Where tentative conclusions are still acceptable, α ≥ .
C69	Alpha levels and beta levels are related: An alpha level is the probability of a type I error, or rejecting the null hypothesis when it is true. A beta level, usually just called beta(β), is the opposite; the probability of of accepting the null hypothesis when it's false.
C70	A commonly used rule says that a data point is an outlier if it is more than 1.5 ⋅ IQR 1.5\cdot \text{IQR} 1. 5⋅IQR1, point, 5, dot, start text, I, Q, R, end text above the third quartile or below the first quartile.
C71	Properties of the SVD U is a n × k matrix with orthonormal columns, UT U = Ik, where Ik is the k × k identity matrix. V is an orthonormal k × k matrix, V T = V −1 . S is a k ×k diagonal matrix, with the non-negative singular values, s1,s2,,sk, on the diagonal.
C72	A random variable is a variable whose value is unknown or a function that assigns values to each of an experiment's outcomes. Random variables are often used in econometric or regression analysis to determine statistical relationships among one another.
C73	For multi class classification using SVM; It is NOT (one vs one) and NOT (one vs REST). Instead learn a two-class classifier where the feature vector is (x, y) where x is data and y is the correct label associated with the data.
C74	One-shot learning is a classification task where one example (or a very small number of examples) is given for each class, that is used to prepare a model, that in turn must make predictions about many unknown examples in the future.
C75	How To Develop a Machine Learning Model From ScratchDefine adequately our problem (objective, desired outputs…).Gather data.Choose a measure of success.Set an evaluation protocol and the different protocols available.Prepare the data (dealing with missing values, with categorial values…).Spilit correctly the data.More items
C76	Unlike linear regression which outputs continuous number values, logistic regression transforms its output using the logistic sigmoid function to return a probability value which can then be mapped to two or more discrete classes.
C77	The main reason why we use sigmoid function is because it exists between (0 to 1). Therefore, it is especially used for models where we have to predict the probability as an output. Since probability of anything exists only between the range of 0 and 1, sigmoid is the right choice. The function is differentiable.
C78	Advantages and Disadvantages of Artificial Intelligence Reduction in Human Error: The phrase “human error” was born because humans make mistakes from time to time. Takes risks instead of Humans: Available 24x7: Helping in Repetitive Jobs: Digital Assistance: Faster Decisions: Daily Applications: New Inventions:
C79	Calculating the distance of various points in the scene relative to the position of the camera is one of the important tasks for a computer vision system.
C80	According to the central limit theorem, the mean of a sampling distribution of means is an unbiased estimator of the population mean. Note that the larger the sample, the less variable the sample mean. The mean of many observations is less variable than the mean of few.
C81	An embedding is a relatively low-dimensional space into which you can translate high-dimensional vectors. Embeddings make it easier to do machine learning on large inputs like sparse vectors representing words. An embedding can be learned and reused across models.
C82	Mini-batch gradient descent is a variation of the gradient descent algorithm that splits the training dataset into small batches that are used to calculate model error and update model coefficients. It is the most common implementation of gradient descent used in the field of deep learning.
C83	Minimax GAN loss refers to the minimax simultaneous optimization of the discriminator and generator models. Minimax refers to an optimization strategy in two-player turn-based games for minimizing the loss or cost for the worst case of the other player.
C84	Reinforcement Learning(RL) is a type of machine learning technique that enables an agent to learn in an interactive environment by trial and error using feedback from its own actions and experiences.
C85	While there are a number of different methods for measuring intelligence, the standard and most widely accepted method is by measuring a person's 'intelligence quotient' or IQ. Based on a series of tests which assess various types of abilities such a mathematical, spatial, verbal, logic and memory.
C86	general Neural Networks end up solving a non-convex optimization methods , while Logistic Regression end up with a Convex Optimization problem for which global optima can be found very efficiently. Modern neural networks rarely use Logistic regression. They use ReLU or tanH as the neural activation function.
C87	The main argument against using linear regression for time series data is that we're usually interested in predicting the future, which would be extrapolation (prediction outside the range of the data) for linear regression. Extrapolating linear regression is seldom reliable.
C88	Bootstrapping assigns measures of accuracy (bias, variance, confidence intervals, prediction error, etc.) to sample estimates. This technique allows estimation of the sampling distribution of almost any statistic using random sampling methods.
C89	3:456:33Suggested clip · 56 secondsStatQuest - Sample Size and Effective Sample Size, Clearly ExplainedYouTubeStart of suggested clipEnd of suggested clip
C90	5 ways to deal with outliers in dataSet up a filter in your testing tool. Even though this has a little cost, filtering out outliers is worth it. Remove or change outliers during post-test analysis. Change the value of outliers. Consider the underlying distribution. Consider the value of mild outliers.
C91	The standard solution that psychologists take to measuring latent variables is to use a series of questions that are all designed to measure the latent variable. This is known as a multi-item scale, where an “item” is a question, and a “scale” is the resulting estimate of the latent variable.
C92	Spreading activation is a method for searching associative networks, biological and artificial neural networks, or semantic networks. Spreading activation can also be applied in information retrieval, by means of a network of nodes representing documents and terms contained in those documents.
C93	Part 1: Making the CalculationsStep 1: Find p,q, and n:Step 2: Figure out if you can use the normal approximation to the binomial. Step 3: Find the mean, μ by multiplying n and p: Step 4: Multiply step 3 by q : Step 5: Take the square root of step 4 to get the standard deviation, σ:More items
C94	The Wilcoxon test is a nonparametric statistical test that compares two paired groups, and comes in two versions the Rank Sum test or the Signed Rank test. The goal of the test is to determine if two or more sets of pairs are different from one another in a statistically significant manner.
C95	DBSCAN works as such: Divides the dataset into n dimensions. For each point in the dataset, DBSCAN forms an n dimensional shape around that data point, and then counts how many data points fall within that shape. DBSCAN counts this shape as a cluster.
C96	They have a similar structure but they apply under different conditions and guarantee different kinds of points. IVT guarantees a point where the function has a certain value between two given values. MVT guarantees a point where the derivative has a certain value.
C97	Mixed effects logistic regression is used to model binary outcome variables, in which the log odds of the outcomes are modeled as a linear combination of the predictor variables when data are clustered or there are both fixed and random effects.
C98	The Monty Hall problem has confused people for decades. In the game show, Let's Make a Deal, Monty Hall asks you to guess which closed door a prize is behind. The answer is so puzzling that people often refuse to accept it! The problem occurs because our statistical assumptions are incorrect.
C99	Random forest (RF) is a machine-learning method that generally works well with high-dimensional problems and allows for nonlinear relationships between predictors; however, the presence of correlated predictors has been shown to impact its ability to identify strong predictors.

End of preview. Expand in Data Studio

MLQuestions

An MTEB dataset

Massive Text Embedding Benchmark

MLQuestions is a domain adaptation dataset for the machine learning domainIt consists of ML questions along with passages from Wikipedia machine learning pages (https://en.wikipedia.org/wiki/Category:Machine_learning)


Task category	t2t
Domains	Encyclopaedic, Academic, Written
Reference	https://github.com/McGill-NLP/MLQuestions

How to evaluate on this task

You can evaluate an embedding model on this dataset using the following code:

import mteb

task = mteb.get_tasks(["MLQuestions"])
evaluator = mteb.MTEB(task)

model = mteb.get_model(YOUR_MODEL)
evaluator.run(model)

To learn more about how to run models on mteb task check out the GitHub repitory.

Citation

If you use this dataset, please cite the dataset as well as mteb, as this dataset likely includes additional processing as a part of the MMTEB Contribution.


@inproceedings{kulshreshtha-etal-2021-back,
  abstract = {In this work, we introduce back-training, an alternative to self-training for unsupervised domain adaptation (UDA). While self-training generates synthetic training data where natural inputs are aligned with noisy outputs, back-training results in natural outputs aligned with noisy inputs. This significantly reduces the gap between target domain and synthetic data distribution, and reduces model overfitting to source domain. We run UDA experiments on question generation and passage retrieval from the Natural Questions domain to machine learning and biomedical domains. We find that back-training vastly outperforms self-training by a mean improvement of 7.8 BLEU-4 points on generation, and 17.6{\%} top-20 retrieval accuracy across both domains. We further propose consistency filters to remove low-quality synthetic data before training. We also release a new domain-adaptation dataset - MLQuestions containing 35K unaligned questions, 50K unaligned passages, and 3K aligned question-passage pairs.},
  address = {Online and Punta Cana, Dominican Republic},
  author = {Kulshreshtha, Devang  and
Belfer, Robert  and
Serban, Iulian Vlad  and
Reddy, Siva},
  booktitle = {Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing},
  month = nov,
  pages = {7064--7078},
  publisher = {Association for Computational Linguistics},
  title = {Back-Training excels Self-Training at Unsupervised Domain Adaptation of Question Generation and Passage Retrieval},
  url = {https://aclanthology.org/2021.emnlp-main.566},
  year = {2021},
}


@article{enevoldsen2025mmtebmassivemultilingualtext,
  title={MMTEB: Massive Multilingual Text Embedding Benchmark},
  author={Kenneth Enevoldsen and Isaac Chung and Imene Kerboua and Márton Kardos and Ashwin Mathur and David Stap and Jay Gala and Wissam Siblini and Dominik Krzemiński and Genta Indra Winata and Saba Sturua and Saiteja Utpala and Mathieu Ciancone and Marion Schaeffer and Gabriel Sequeira and Diganta Misra and Shreeya Dhakal and Jonathan Rystrøm and Roman Solomatin and Ömer Çağatan and Akash Kundu and Martin Bernstorff and Shitao Xiao and Akshita Sukhlecha and Bhavish Pahwa and Rafał Poświata and Kranthi Kiran GV and Shawon Ashraf and Daniel Auras and Björn Plüster and Jan Philipp Harries and Loïc Magne and Isabelle Mohr and Mariya Hendriksen and Dawei Zhu and Hippolyte Gisserot-Boukhlef and Tom Aarsen and Jan Kostkan and Konrad Wojtasik and Taemin Lee and Marek Šuppa and Crystina Zhang and Roberta Rocca and Mohammed Hamdy and Andrianos Michail and John Yang and Manuel Faysse and Aleksei Vatolin and Nandan Thakur and Manan Dey and Dipam Vasani and Pranjal Chitale and Simone Tedeschi and Nguyen Tai and Artem Snegirev and Michael Günther and Mengzhou Xia and Weijia Shi and Xing Han Lù and Jordan Clive and Gayatri Krishnakumar and Anna Maksimova and Silvan Wehrli and Maria Tikhonova and Henil Panchal and Aleksandr Abramov and Malte Ostendorff and Zheng Liu and Simon Clematide and Lester James Miranda and Alena Fenogenova and Guangyu Song and Ruqiya Bin Safi and Wen-Ding Li and Alessia Borghini and Federico Cassano and Hongjin Su and Jimmy Lin and Howard Yen and Lasse Hansen and Sara Hooker and Chenghao Xiao and Vaibhav Adlakha and Orion Weller and Siva Reddy and Niklas Muennighoff},
  publisher = {arXiv},
  journal={arXiv preprint arXiv:2502.13595},
  year={2025},
  url={https://arxiv.org/abs/2502.13595},
  doi = {10.48550/arXiv.2502.13595},
}

@article{muennighoff2022mteb,
  author = {Muennighoff, Niklas and Tazi, Nouamane and Magne, Lo{\"\i}c and Reimers, Nils},
  title = {MTEB: Massive Text Embedding Benchmark},
  publisher = {arXiv},
  journal={arXiv preprint arXiv:2210.07316},
  year = {2022}
  url = {https://arxiv.org/abs/2210.07316},
  doi = {10.48550/ARXIV.2210.07316},
}

Dataset Statistics

The following code contains the descriptive statistics from the task. These can also be obtained using:

import mteb

task = mteb.get_task("MLQuestions")

desc_stats = task.metadata.descriptive_stats

{
    "dev": {
        "num_samples": 12500,
        "number_of_characters": 2915233,
        "num_documents": 11000,
        "min_document_length": 3,
        "average_document_length": 258.8772727272727,
        "max_document_length": 395,
        "unique_documents": 11000,
        "num_queries": 1500,
        "min_query_length": 14,
        "average_query_length": 45.05533333333333,
        "max_query_length": 160,
        "unique_queries": 1500,
        "none_queries": 0,
        "num_relevant_docs": 1500,
        "min_relevant_docs_per_query": 1,
        "average_relevant_docs_per_query": 1.0,
        "max_relevant_docs_per_query": 1,
        "unique_relevant_docs": 1500,
        "num_instructions": null,
        "min_instruction_length": null,
        "average_instruction_length": null,
        "max_instruction_length": null,
        "unique_instructions": null,
        "num_top_ranked": null,
        "min_top_ranked_per_query": null,
        "average_top_ranked_per_query": null,
        "max_top_ranked_per_query": null
    },
    "test": {
        "num_samples": 12500,
        "number_of_characters": 2916280,
        "num_documents": 11000,
        "min_document_length": 3,
        "average_document_length": 258.8772727272727,
        "max_document_length": 395,
        "unique_documents": 11000,
        "num_queries": 1500,
        "min_query_length": 12,
        "average_query_length": 45.75333333333333,
        "max_query_length": 165,
        "unique_queries": 1500,
        "none_queries": 0,
        "num_relevant_docs": 1500,
        "min_relevant_docs_per_query": 1,
        "average_relevant_docs_per_query": 1.0,
        "max_relevant_docs_per_query": 1,
        "unique_relevant_docs": 1499,
        "num_instructions": null,
        "min_instruction_length": null,
        "average_instruction_length": null,
        "max_instruction_length": null,
        "unique_instructions": null,
        "num_top_ranked": null,
        "min_top_ranked_per_query": null,
        "average_top_ranked_per_query": null,
        "max_top_ranked_per_query": null
    }
}

This dataset card was automatically generated using MTEB

Downloads last month: 54

Size of downloaded dataset files:

3.52 MB

Size of the auto-converted Parquet files:

3.52 MB

Number of rows:

28,000