MCTI Recommendation Task (uncased) DRAFT
Disclaimer: The Brazilian Ministry of Science, Technology, and Innovation (MCTI) has partially supported this project.
The model NLP MCTI Recommendation Multi is part of the project Research Financing Product Portfolio (FPP) focuses on the task of Recommendation and explores different machine learning strategies that provide suggestions of items that are likely to be handy for a particular individual. Several methods were faced against each other to compare the error estimatives. Using LDA model, a simulated dataset was created.
According to the abstract,
Current model card disposes model's description and it's classes. Also, inteded uses are described along with a "how to use" section, exposing necessary conditions for the data used. Further in the card, data and it's limitation and bias were discussed. Tables along the page supports the information and tests that were made. How the recommendation is made, datasets used and the benchmarks generated are all set all over the model card.
Model description
The surprise library provides 11 classifier models that try to predict the classification of training data based on several different collaborative-filtering techniques. The models provided with a brief explanation in English are mentioned below, for more information please refer to the package documentation.
random_pred.NormalPredictor: Algorithm predicting a random rating based on the distribution of the training set, which is assumed to be normal.
baseline_only.BaselineOnly: Algorithm predicting the baseline estimate for given user and item.
knns.KNNBasic: A basic collaborative filtering algorithm.
knns.KNNWithMeans: A basic collaborative filtering algorithm, taking into account the mean ratings of each user.
knns.KNNWithZScore: A basic collaborative filtering algorithm, taking into account the z-score normalization of each user.
knns.KNNBaseline: A basic collaborative filtering algorithm taking into account a baseline rating.
matrix_factorization.SVD: The famous SVD algorithm, as popularized by Simon Funk during the Netflix Prize.
matrix_factorization.SVDpp: The SVD++ algorithm, an extension of SVD taking into account implicit ratings.
matrix_factorization.NMF: A collaborative filtering algorithm based on Non-negative Matrix Factorization.
slope_one.SlopeOne: A simple yet accurate collaborative filtering algorithm.
co_clustering.CoClustering: A collaborative filtering algorithm based on co-clustering.
It is possible to pass a custom dataframe as an argument to this class. The dataframe in question needs to have 3 columns with the following name: ['userID', 'itemID', 'rating'].
class Method:
def __init__(self,df):
self.df=df
self.available_methods=[
'surprise.NormalPredictor',
'surprise.BaselineOnly',
'surprise.KNNBasic',
'surprise.KNNWithMeans',
'surprise.KNNWithZScore',
'surprise.KNNBaseline',
'surprise.SVD',
'surprise.SVDpp',
'surprise.NMF',
'surprise.SlopeOne',
'surprise.CoClustering',
]
def show_methods(self):
print('The avaliable methods are:')
for i,method in enumerate(self.available_methods):
print(str(i)+': '+method)
def run(self,the_method):
self.the_method=the_method
if(self.the_method[0:8]=='surprise'):
self.run_surprise()
elif(self.the_method[0:6]=='Gensim'):
self.run_gensim()
elif(self.the_method[0:13]=='Transformers-'):
self.run_transformers()
else:
print('This method is not defined! Try another one.')
def run_surprise(self):
from surprise import Reader
from surprise import Dataset
from surprise.model_selection import train_test_split
reader = Reader(rating_scale=(1, 5))
data = Dataset.load_from_df(self.df[['userID', 'itemID', 'rating']], reader)
trainset, testset = train_test_split(data, test_size=.30)
the_method=self.the_method.replace("surprise.", "")
eval(f"exec('from surprise import {the_method}')")
the_algorithm=locals()[the_method]()
the_algorithm.fit(trainset)
self.predictions=the_algorithm.test(testset)
list_predictions=[(uid,iid,r_ui,est) for uid,iid,r_ui,est,_ in self.predictions]
self.predictions_df = pd.DataFrame(list_predictions, columns =['user_id', 'item_id', 'rating','predicted_rating'])
Every model was used and evaluated. When faced with each other different methods presented different error estimatives.
The surprise library provides 4 different methods to assess the accuracy of the ratings prediction. Those are: rmse, mse, mae and fcp. For further discussion on each metric please visit the package documentation.
class Evaluator:
def __init__(self,predictions_df):
self.available_evaluators=['surprise.rmse','surprise.mse',
'surprise.mae','surprise.fcp']
self.predictions_df=predictions_df
def show_evaluators(self):
print('The avaliable evaluators are:')
for i,evaluator in enumerate(self.available_evaluators):
print(str(i)+': '+evaluator)
def run(self,the_evaluator):
self.the_evaluator=the_evaluator
if(self.the_evaluator[0:8]=='surprise'):
self.run_surprise()
else:
print('This evaluator is not available!')
def run_surprise(self):
import surprise
from surprise import accuracy
predictions=[surprise.prediction_algorithms.predictions.Prediction(row['user_id'],row['item_id'],row['rating'],row['predicted_rating'],{}) for index,row in self.predictions_df.iterrows()]
self.predictions=predictions
self.the_evaluator= 'accuracy.' + self.the_evaluator.replace("surprise.", "")
self.acc = eval(f'{self.the_evaluator}(predictions,verbose=True)')
Intended uses
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to be fine-tuned on a downstream task. See the model hub to look for fine-tuned versions of a task that interests you. Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked) to make decisions, such as sequence classification, token classification or question answering. For tasks such as text generation you should look at model like XXX.
How to use
The datasets for collaborative filtering must be: - The dataframe containing the ratings. - It must have three columns, corresponding to the user (raw) ids, the item (raw) ids, and the ratings, in this order.
>>> import pandas as pd
>>> import numpy as np
class Data:
The databases (ml_100k, ml_1m and jester) are built-in the surprise package for collaborative-filtering.
def_init_(self):
self.available_databases=['ml_100k', 'ml_1m','jester', 'lda_topics', 'lda_rankings', 'uniform']
def show_available_databases(self):
print('The avaliable database are:')
for i,database in enumerate(self.available_databases):
print(str(i)+': '+database)
def read_data(self,database_name):
self.database_name=database_name
self.the_data_reader= getattr(self, 'read_'+database_name.lower())
self.the_data_reader()
def read_ml_100k(self):
from surprise import Dataset
data = Dataset.load_builtin('ml-100k')
self.df = pd.DataFrame(data.__dict__['raw_ratings'], columns=['user_id','item_id','rating','timestamp'])
self.df.drop(columns=['timestamp'],inplace=True)
self.df.rename({'user_id':'userID','item_id':'itemID'},axis=1,inplace=True)
def read_ml_1m(self):
from surprise import Dataset
data = Dataset.load_builtin('ml-1m')
self.df = pd.DataFrame(data.__dict__['raw_ratings'], columns=['user_id','item_id','rating','timestamp'])
self.df.drop(columns=['timestamp'],inplace=True)
self.df.rename({'user_id':'userID','item_id':'itemID'},axis=1,inplace=True)
def read_jester(self):
from surprise import Dataset
data = Dataset.load_builtin('jester')
self.df = pd.DataFrame(data.__dict__['raw_ratings'], columns=['user_id','item_id','rating','timestamp'])
self.df.drop(columns=['timestamp'],inplace=True)
self.df.rename({'user_id':'userID','item_id':'itemID'},axis=1,inplace=True)
Hyperparameters -
n_users : number of simulated users in the database;
n_ratings : number of simulated rating events in the database.
This is a fictional dataset based in the choice of an uniformly distributed random rating(from 1 to 5) for one of the simulated users of the recommender-system that is being designed in this research project.
def read_uniform(self):
n_users = 20
n_ratings = 10000
import random
opo = pd.read_csv('../oportunidades.csv')
df = [(random.randrange(n_users), random.randrange(len(opo)), random.randrange(1,5)) for i in range(n_ratings)]
self.df = pd.DataFrame(df, columns = ['userID', 'itemID', 'rating'])
Hyperparameters -
n_users` : number of simulated users in the database;
n_ratings` : number of simulated rating events in the database.
This first LDA based dataset builds a model with K = n_users topics. LDA topics are used as proxies for simulated users with different clusters of interest. At first a random opportunity is chosen, than the amount of a randomly chosen topic inside the description is multiplied by five. The ceiling operation of this result is the rating that the fictional user will give to that opportunity. Because the amount of each topic predicted by the model is disollved among various topics, it is very rare to find an opportunity that has a higher LDA value. The consequence is that this dataset has really low volatility and the major part of ratings are equal to 1.
def read_lda_topics(self):
n_users = 20
n_ratings = 10000
import gensim
import random
import math
opo = pd.read_csv('../oportunidades_results.csv')
# opo = opo.iloc[np.where(opo['opo_brazil']=='Y')]
try:
lda_model = gensim.models.ldamodel.LdaModel.load(f'models/lda_model{n_users}.model')
except:
import generate_users
generate_users.gen_model(n_users)
lda_model = gensim.models.ldamodel.LdaModel.load(f'models/lda_model{n_users}.model')
df = []
for i in range(n_ratings):
opo_n = random.randrange(len(opo))
txt = opo.loc[opo_n,'opo_texto']
opo_bow = lda_model.id2word.doc2bow(txt.split())
topics = lda_model.get_document_topics(opo_bow)
topics = {topic[0]:topic[1] for topic in topics}
user = random.sample(topics.keys(), 1)[0]
rating = math.ceil(topics[user]*5)
df.append((user, opo_n, rating))
self.df = pd.DataFrame(df, columns = ['userID', 'itemID', 'rating'])
def read_lda_rankings(self):
n_users = 9
n_ratings = 1000
import gensim
import random
import math
import tqdm
opo = pd.read_csv('../oportunidades.csv')
opo = opo.iloc[np.where(opo['opo_brazil']=='Y')]
opo.index = range(len(opo))
path = f'models/output_linkedin_cle_lda_model_{n_users}_topics_symmetric_alpha_auto_beta'
lda_model = gensim.models.ldamodel.LdaModel.load(path)
df = []
pbar = tqdm.tqdm(total= n_ratings)
for i in range(n_ratings):
opo_n = random.randrange(len(opo))
txt = opo.loc[opo_n,'opo_texto']
opo_bow = lda_model.id2word.doc2bow(txt.split())
topics = lda_model.get_document_topics(opo_bow)
topics = {topic[0]:topic[1] for topic in topics}
prop = pd.DataFrame([topics], index=['prop']).T.sort_values('prop', ascending=True)
prop['rating'] = range(1, len(prop)+1)
prop['rating'] = prop['rating']/len(prop)
prop['rating'] = prop['rating'].apply(lambda x: math.ceil(x*5))
prop.reset_index(inplace=True)
prop = prop.sample(1)
df.append((prop['index'].values[0], opo_n, prop['rating'].values[0]))
pbar.update(1)
pbar.close()
self.df = pd.DataFrame(df, columns = ['userID', 'itemID', 'rating'])
Limitations and bias
In this model we have faced some obstacles that we had overcome, but some of those, by the nature of the project, couldn't be totally solved. Databases containing profiles of possible users of the planned prototype are not available. For this reason, it was necessary to carry out simulations in order to represent the interests of these users, so that the recommendation system could be modeled. A simulation of clusters of latent interests was realized, based on topics present in the texts describing financial products. Due the fact that the dataset was build it by ourselves, there was no interaction yet between a user and the dataset, therefore we don't have realistic ratings, making the results less believable.
Later on, we have used a database of scrappings of linkedin profiles. The problem is that the profiles that linkedin shows is biased, so the profiles that appears was geographically closed, or related to the users organization and email.
Training data
To train the Latent Dirichlet allocation (LDA) model, it was used a database of a scrapping of Researchers profiles on Linkedin.
Training procedure
Evaluation results
Checkpoints
- Example
data=Data()
data.show_available_databases()
data.read_data('ml_100k')
method=Method(data.df)
method.show_methods()
method.run('surprise.KNNWithMeans')
predictions_df=method.predictions_df
evaluator=Evaluator(predictions_df)
evaluator.show_evaluators()
evaluator.run('surprise.mse')
The avaliable database are: 0: ml_100k
1: ml_1m
2: jester
3: lda_topics
4: lda_rankings
5: uniform
The avaliable methods are:
0: surprise.NormalPredictor
1: surprise.BaselineOnly
2: surprise.KNNBasic
3: surprise.KNNWithMeans
4: surprise.KNNWithZScore
5: surprise.KNNBaseline
6: surprise.SVD
7: surprise.SVDpp
8: surprise.NMF
9: surprise.SlopeOne
10: surprise.CoClustering
Computing the msd similarity matrix...
Done computing similarity matrix.
The avaliable evaluators are:
0: surprise.rmse
1: surprise.mse
2: surprise.mae
3: surprise.fcp
MSE: 0.9146
Next, we have the code that builds the table with the accuracy metrics for all rating prediction models built-in the surprise package. The expected return of this function is a pandas dataframe (11x4) corresponding to the 11 classifier models and 4 different accuracy metrics.
def model_table(label):
import tqdm
table = pd.DataFrame()
data=Data()
data.read_data(label)
method=Method(data.df)
for m in method.available_methods:
print(m)
method.run(m)
predictions_df=method.predictions_df
evaluator=Evaluator(predictions_df)
metrics = []
for e in evaluator.available_evaluators:
evaluator.run(e)
metrics.append(evaluator.acc)
table = table.append(dict(zip(evaluator.available_evaluators,metrics)),ignore_index=True)
table.index = [x[9:] for x in method.available_methods]
table.columns = [x[9:].upper() for x in evaluator.available_evaluators]
return table
import sys, os
sys.stdout = open(os.devnull, 'w') # Codigo para desativar os prints
uniform = model_table('uniform')
#topics = model_table('lda_topics')
ranking = model_table('lda_rankings')
sys.stdout = sys.__stdout__ # Codigo para reativar os prints
- Usage Example
In this section it will be explained how the recommendation is made for the user.
import gradio as gr
import random
import pandas as pd
opo = pd.read_csv('oportunidades_results.csv', lineterminator='\n')
# opo = opo.iloc[np.where(opo['opo_brazil']=='Y')]
simulation = pd.read_csv('simulation2.csv')
userID = max(simulation['userID']) + 1
This function, creates the string that it will be displayed to the user on the app, showing the opportunities title, link and the resume.
def build_display_text(opo_n):
title = opo.loc[opo_n]['opo_titulo']
link = opo.loc[opo_n]['link']
summary = opo.loc[opo_n]['facebook-bart-large-cnn_results']
display_text = f"**{title}**\n\nURL:\n{link}\n\nSUMMARY:\n{summary}"
return display_text
Here it will be generate 4 random opportunities.
opo_n_one = random.randrange(len(opo))
opo_n_two = random.randrange(len(opo))
opo_n_three = random.randrange(len(opo))
opo_n_four = random.randrange(len(opo))
evaluated = []
The next function, is the "predict_next", that accepts an option and a rating.
def predict_next(option, nota):
global userID
global opo_n_one
global opo_n_two
global opo_n_three
global opo_n_four
global evaluated
global opo
global simulation
Here it will be taken the number, on our database, of the rated opportunity.
selected = [opo_n_one, opo_n_two, opo_n_three, opo_n_four][int(option)-1]
Here is created a new database called simulation, that takes the previous simulation then adds a new line with te ID of the user, the rated item and the rate. integrates the selected opportunity.
simulation = simulation.append({'userID': userID, 'itemID': selected, 'rating': nota}, ignore_index=True)
evaluated.append(selected)
from surprise import Reader
reader = Reader(rating_scale=(1, 5))
from surprise import Dataset
data = Dataset.load_from_df(simulation[['userID', 'itemID', 'rating']], reader)
trainset = data.build_full_trainset()
from surprise import SVDpp
svdpp = SVDpp()
svdpp.fit(trainset)
items = list()
est = list()
for i in range(len(opo)):
if i not in evaluated:
items.append(i)
est.append(svdpp.predict(userID, i).est)
opo_n_one = items[est.index(sorted(est)[-1])]
opo_n_two = items[est.index(sorted(est)[-2])]
opo_n_three = items[est.index(sorted(est)[-3])]
opo_n_four = items[est.index(sorted(est)[-4])]
return build_display_text(opo_n_one), build_display_text(opo_n_two), build_display_text(opo_n_three), build_display_text(opo_n_four)
Here we have the interation of gradio, that allows the construction of the app.
with gr.Blocks() as demo:
with gr.Row():
one_opo = gr.Textbox(build_display_text(opo_n_one), label='Oportunidade 1')
two_opo = gr.Textbox(build_display_text(opo_n_two), label='Oportunidade 2')
with gr.Row():
three_opo = gr.Textbox(build_display_text(opo_n_three), label='Oportunidade 3')
four_opo = gr.Textbox(build_display_text(opo_n_four), label='Oportunidade 4')
with gr.Row():
option = gr.Radio(['1', '2', '3', '4'], label='Opção', value = '1')
with gr.Row():
nota = gr.Slider(1,5,step=1,label="Nota 1")
with gr.Row():
confirm = gr.Button("Confirmar")
confirm.click(fn=predict_next,
inputs=[option, nota],
outputs=[one_opo, two_opo, three_opo, four_opo])
if __name__ == "__main__":
demo.launch()
Benchmarks
# LDA-GENERATED DATASET
ranking
| RMSE | MSE | MAE | FCP | |
|---|---|---|---|---|
| NormalPredictor | 1.820737 | 3.315084 | 1.475522 | 0.514134 |
| BaselineOnly | 1.072843 | 1.150992 | 0.890233 | 0.556560 |
| KNNBasic | 1.232248 | 1.518436 | 0.936799 | 0.648604 |
| KNNWithMeans | 1.124166 | 1.263750 | 0.808329 | 0.597148 |
| KNNWithZScore | 1.056550 | 1.116299 | 0.750004 | 0.669651 |
| KNNBaseline | 1.134660 | 1.287454 | 0.825161 | 0.614270 |
| SVD | 0.977468 | 0.955444 | 0.757485 | 0.723829 |
| SVDpp | 0.843065 | 0.710758 | 0.670516 | 0.671737 |
| NMF | 1.122684 | 1.260420 | 0.722101 | 0.688728 |
| SlopeOne | 1.073552 | 1.152514 | 0.747142 | 0.651937 |
| CoClustering | 1.293383 | 1.672838 | 1.007951 | 0.494174 |
# BENCHMARK DATASET
uniform
| RMSE | MSE | MAE | FCP | |
|---|---|---|---|---|
| NormalPredictor | 1.508925 | 2.276854 | 1.226758 | 0.503723 |
| BaselineOnly | 1.153331 | 1.330172 | 1.022732 | 0.506818 |
| KNNBasic | 1.205058 | 1.452165 | 1.026591 | 0.501168 |
| KNNWithMeans | 1.202024 | 1.444862 | 1.028149 | 0.503527 |
| KNNWithZScore | 1.216041 | 1.478756 | 1.041070 | 0.501582 |
| KNNBaseline | 1.225609 | 1.502117 | 1.048107 | 0.498198 |
| SVD | 1.176273 | 1.383619 | 1.013285 | 0.502067 |
| SVDpp | 1.192619 | 1.422340 | 1.018717 | 0.500909 |
| NMF | 1.338216 | 1.790821 | 1.120604 | 0.492944 |
| SlopeOne | 1.224219 | 1.498713 | 1.047170 | 0.494298 |
| CoClustering | 1.223020 | 1.495778 | 1.033699 | 0.518509 |
BibTeX entry and citation info
@unpublished{recommend22,
author ={Jo\~{a}o Gabriel de Moraes Souza. and Daniel Oliveira Cajueiro. and Johnathan de O. Milagres. and Vin\´{i}cius de Oliveira Watanabe. and V\´{i}tor Bandeira Borges. and Victor Rafael Celestino.},
title ={A comprehensive review of recommendation systems: method, data, evaluation and coding},
}
