Comparison of Output Tokens per Second; Higher is better by Provider

This dataset belongs to a single product on Amazon and contains detailed information about its reviews. Each entry in the dataset represents a review written by a customer.

reviewerid: A unique identifier for the reviewer. Each reviewer has a distinct ID that helps differentiate their reviews from others. asin: Amazon Standard Identification Number. It is a unique identifier assigned to each product on Amazon. reviewername: The name or username of the reviewer. This is the display name of the person who wrote the review. helpful: A measure indicating how many people found the review helpful. This is often shown as a ratio, e.g., [2,3] where 2 people found it helpful out of 3 total votes. reviewtext: The actual text of the review. This is the content of what the reviewer wrote about the product. overall: The rating given by the reviewer, usually on a scale of 1 to 5 stars. summary: A short summary or title of the review. This is often a brief highlight of the reviewer's opinion. unixreviewtime: The time the review was written, represented as a Unix timestamp. This is the number of seconds that have elapsed since January 1, 1970 (midnight UTC/GMT). reviewtime: The human-readable date when the review was written, typically in the format "MM DD, YYYY". day_diff: The difference in days between the review date and some reference date (often the current date or the date the dataset was compiled). This helps to understand how recent the review is. helpful_yes: The number of people who found the review helpful. This is the first number in the "helpful" ratio. total_vote: The total number of votes the review received. This is the second number in the "helpful" ratio.

Based on a survey conducted in Germany in 2020, there was no significant difference among age groups when it came to using Amazon to buy or sell second-hand items online. Among respondents aged between 16 and 29 years, ** percent said that they had used Amazon for this purpose before, while ** percent of respondents aged 30 years and older said the same.

This dataset contains meticulously cleaned and structured web traffic data collected across multiple websites, including Amazon platforms and services like Amazon Prime, AWS, and AWS Support. It spans various traffic sources, user devices, key actions, and engagement metrics, making it a powerful resource for digital marketing analysis, customer behavior modeling, and time series forecasting.

Ideal for:

Web traffic analysis Conversion rate optimization Bounce rate analysis User segmentation Predictive modeling and machine learning 📌 Dataset Features: Rows: 2006 Columns: 18

Date Range: Starts from January 1st, 2019 (Exact end date can be inferred from the dataset)

🔍 Columns Overview: Country: Country of user origin

Timestamp: Full timestamp of the visit Device Category: Type of device (Desktop, Mobile, Tablet) Key Actions: User actions like Purchase, Sign Up, Subscribe Page Path: Visited page (e.g., /home, /contact) Source: Traffic source (e.g., organic search, social media) Avg Session Duration: Duration of session in seconds Bounce Rate: % of single-page sessions Conversions: Number of conversions New Users: Number of new users in session Page Views: Total page views Returning Users: Count of returning users Unique Page Views: Unique page views Average time on home page (min): Self-explanatory Website: Name of the specific Amazon service or domain Date, Time, Day: Parsed date and time information

📊 Potential Use Cases: Machine Learning: Predicting bounce rate, conversion likelihood, or segmenting user behavior. Business Intelligence: Dashboards for performance analysis by device, source, or day. Time Series Forecasting: Analyze traffic patterns over time. A/B Testing: Benchmarking traffic changes across page paths or traffic sources.

The dataset consists of a collection of reviews from the Booking website, during the period between June 2016 and August 2016. The second dataset includes reviews taken from the Amazon website and it is a subset of the dataset available at http://jmcauley.ucsd.edu/data/amazon, previously used in [14,15].

Comprehensive comparison of Latency (Time to First Token) vs. Output Speed (Output Tokens per Second) by Provider

Summary

Time-series (2018-2024) of the Amazon dashboard, including minor updates to the methods.

The Amazon dashboard data product tracks individual fire events across most of South America (10N - 25S, 85W - 30W) in near-real time. The model classifies fires into four key fire types (deforestation, forest, small clearing and agricultural, and savanna and grassland fires) and provides estimates of individual fire carbon emissions. Methods are described in Andela et al. (2022). Near-real time estimates are provided at https://amzfire.servirglobal.net/ and here we archive historic time-series.

Methods

The data archived here (v1.1) include several small updates.

Two updates relate to the use of VIIRS active fire detections. First, VIIRS active fire detections have been updated from collection 1 to collection 2. Second, any full day of missing data from either the VIIRS instrument onboard NOAA-20 or Suomi NPP is now replaced by data of the other instrument. This "gap" filling helps reduce the impact of periods with instrument outage, like those of Suomi NPP VIIRS during the 2024 burning season.

The other two updates relate to the emissions calculations. First, to convert dry matter burned to carbon emissions, we have introduced fire type specific emissions factors instead of the earlier assumption of 50% carbon content for all fire types. Second, as part of the Sense4Fire project (https://sense4fire.eu/), we provide daily gridded emissions estimates of Dry Matter (DM), C, CO2, CO, and NOx at 0.1 degree resolution. We used emissions factors provided by Andrea (2019) for savanna and grassland fires as well as small clearing and agricultural fires while for forest and deforestation fires we reviewed the literature to select the most relevant emissions factors (Table 1).

Table 1: Emissions factors (gram species per kg dry matter burned) used to calculate C, CO2, CO, and NOx emissions.

Dataset description

For full detail, please see Andela et al. (2022). The tables below (Tables 2 - 4) describe the content of the fire event (polygon) and active fire detections (point) shapefiles as well as the gridded emissions product. The active fire detections and associated estimates of dry matter burned can be combined with emissions factors (Table 1) to derive daily trace gas emissions time series for species and areas of interest.

Table 2: Explanation of fire event shapefile attribute table.

Table 3: Explanation of active fire detection shapefile attribute table.

Table 4: Content of daily gridded (0.1 degree resolution) emissions netcdf files. The daily emissions product provides emissions estimates of dry matter, C, CO2, CO, and NOx. For DM and CO partitioned emissions are also provided by fire type, for other species these can be derived by multiplying the dry matter burned (DM) estimates with trace gas specific emissions factors (Table 1). Values of each grid cell can be multiplied by the number of seconds per day and grid cell area to calculate total emissions (convert "kg species m-2 s-1" to "kg species day-1 per grid cell").

Results

Despite the various small improvements to the dataset, the data are largely consistent with the original dataset published for 2019-2020 (Table 5).

Table 5: Comparison of model versions (original from Andela et al., 2022 and v1.1 published here) for April-December 2019 (equator-25S, 85W - 30W). Note that the current version (v1.1) is complete for 2019, but the original dataset had missing data due to incomplete active fire detections from NOAA-20 VIIRS at that time.

Acknowledgements

The Sense4Fire project is funded by ESA under ESA Contract Number: 4000134840/21/I-NB.

References

Andela, N., Morton, D.C., Schroeder, W., Chen, Y., Brando, P.M. and Randerson, J.T., 2022. Tracking and classifying Amazon fire events in near real time. Science advances, 8, eabd2713. https://doi.org/10.1126/sciadv.abd2713.

Andreae, M.O., 2019. Emission of trace gases and aerosols from biomass burning–an updated assessment. Atmospheric Chemistry and Physics, 19, 8523-8546. https://doi.org/10.5194/acp-19-8523-2019.

Comprehensive comparison of Output Speed (Output Tokens per Second) vs. Price (USD per 1M Tokens) by Provider

Amazon Best Seller Dataset

Data from Amazon US on the top 50 selling products for all 40 listed categories. Includes: - Product Name - Product Categories - Raking of product in its category - Product Review Count (Number of Reviews) - Product Review (First page of reviews) - Includes Meta + Full Review + Rating given by review - Cost of item at the time of scraping. - Product URL

Data scraped using Python Selenium and cleaned manually Scraping Notebook

Data Scraping History

PS Dataset will be updated every second week of the month for the time period of 2022 August -> 2023 June

With 438 billion U.S. dollars in net sales, the United States were Amazon’s biggest market in 2024. Germany was ranked second with 41 billion U.S. dollars, ahead of the UK with 37.9 billion U.S. dollars. Biggest internet company Founded in 1994, Amazon has grown into one of the biggest e-commerce marketplaces and cloud computing platforms worldwide. In 2020, Amazon was ranked first in terms of company revenue among global publicly traded internet companies. With an annual revenue of approximately 386 billion U.S. dollars, the e-retailer ranked far ahead of closest competitors Google (181.7 billion U.S. dollars) and Alibaba (109.5 billion U.S. dollars). Amazon shopping Amazon is the leading e-retailer in the United States. In September 2021, 65 percent of Amazon customers in the United States held a membership with the company’s subscription service Amazon Prime, benefitting from free 2-day shipping, music and video streaming and exclusive offers and deals. Amazon Prime members are very engaged shoppers: a June 2021 survey found that over four in ten of them were likely to shop not just during Prime days, but also during other retail events, with 56 percent of them showing interest for Cyber Monday.

According to the source, in the first quarter of 2023, Amazon Prime had a 30-day trial after which ** percent of users subscribed to the service. The conversion rate has increased, as it was ** percent in the same period of 2022. Moreover, ** percent of Amazon Prime members renewed their membership for a year, and ** percent renewed it for a second year over the first three months of 2023.

Business Problem Background

We have a client who has a website where people write different reviews for technical products. Now they are adding a new feature to their website i.e. The reviewer will have to add stars(rating) as well with the review. The rating is out 5 stars and it only has 5 options available 1 star, 2 stars, 3 stars, 4 stars, 5 stars. Now they want to predict ratings for the reviews which were written in the past and they don’t have a rating. So, we have to build an application which can predict the rating by seeing the review.

• A recent survey (Hinckley, 2015) revealed that 67.7% of consumers are effectively influenced by online reviews when making their purchase decisions. More precisely, 54.7% recognized that these reviews were either fairly, very or absolutely important in their purchase decision making. Relying on online reviews has thus become a second nature for consumers.

• **Consumers want to find useful information as quickly as possible. However, searching and comparing text reviews can be frustrating for users as they feel submerged with information. **

• The overall star ratings of the product reviews may not capture the exact polarity of the sentiments. This makes rating prediction a hard problem as customers may assign different ratings for a particular review. In addition, reviews may contain anecdotal information, which do not provide any helpful information and complicates the predictive task.

• **The question that arises is how to successfully predict a user’s numerical rating from its review text content. **

Dataset Information

Data is collected from Amazon.in and flipkart.com using selenium and saved in CSV file. Around 50000 Reviews are collected for this project.

• We have scrape around 50000 reviews of different products along with ratings from amazon website. Basically, we have these columns in dataset. 1) reviews of the product. 2) rating of the product.

• You can fetch other data as well, if you think data can be useful or can help in the project. It completely depends on your imagination or assumption.

Model Building Phase

• **You need to build a machine learning model. **

• This is multi-classification problem and Rating is our target feature class to be predicated in this project. There are five different categories in feature target i.e., The rating is out 5 stars and it only has 5 options available 1 star, 2 stars, 3 stars, 4 stars, 5 stars.

• Try different models with different hyper parameters and select the best model.

The apparel market worldwide in 2024 was led by women's apparel, which had the highest average volume per capita at approximately ***** pieces. Ranked second, children's apparel recorded around **** pieces, followed by men's apparel with about **** pieces.

There is increasing interest in the restoration/regeneration of degraded tropical habitats yet the potential role of natural regenerators remains unclear. We test the hypothesis that the richness and functional diversity of terrestrial mammals differs between forest regrowth stages. We quantified the richness and functional diversity of eight terrestrial mammal seed-disperser species across a forest regrowth gradient in the eastern Brazilian Amazon. We installed camera-traps in 15 sites within small-holder properties with forest regrowth stage classified into three groups, with five sites each of: late second-regrowth forest, early second-regrowth forest and abandoned pasture. Species richness and functional dispersion from the regrowth sites were compared with 15 paired forest control sites. Multi model selection showed that regrowth class was more important for explaining patterns in richness and functional diversity than other variables from three non-mutually exclusive hypotheses: hunting (distance to house, distance to river, distance to town, small holder residence), land cover (% forest cover within 50 meters, 1 kilometer and 5 kilometers) and land use (regrowth class, time since last use). Differences in functional diversity were most strongly explained by a loss of body mass. We found that diversity in regrowth sites could be similar to control sites even in some early-second regrowth areas. This finding suggests that when surrounded by large intact forest areas the richness and functional diversity close to human small-holdings can return to pre-degradation values. Yet we also found a significant reduction in richness and functional diversity in more intensely degraded pasture sites. This reduction in richness and functional diversity may limit the potential for regeneration and increase costs for ecological regeneration and restoration actions around more intense regrowth areas.

Climate, species composition, and soils are thought to control carbon cycling and forest structure in Amazonian forests. Here, we add a demographics scheme (tree recruitment, growth, and mortality) to a recently developed non-demographic model—the Trait-based Forest Simulator (TFS)—to explore the roles of climate and plant traits in controlling forest productivity and structure. We compared two sites with differing climates (seasonal vs. aseasonal precipitation) and plant traits. Through an initial validation simulation, we assessed whether the model converges on observed forest properties (productivity, demographic and structural variables) using datasets of functional traits, structure, and climate to model the carbon cycle at the two sites. In a second set of simulations, we tested the relative importance of climate and plant traits for forest properties within the TFS framework using the climate from the two sites with hypothetical trait distributions representing two axes of functional variation (“fast” vs. “slow” leaf traits, and high vs. low wood density). The adapted model with demographics reproduced observed variation in gross (GPP) and net (NPP) primary production, and respiration. However, NPP and respiration at the level of plant organs (leaf, stem, and root) were poorly simulated. Mortality and recruitment rates were underestimated. The equilibrium forest structure differed from observations of stem numbers suggesting either that the forests are not currently at equilibrium or that mechanisms are missing from the model. Findings from the second set of simulations demonstrated that differences in productivity were driven by climate, rather than plant traits. Contrary to expectation, varying leaf traits had no influence on GPP. Drivers of simulated forest structure were complex, with a key role for wood density mediated by its link to tree mortality. Modeled mortality and recruitment rates were linked to plant traits alone, drought-related mortality was not accounted for. In future, model development should focus on improving allocation, mortality, organ respiration, simulation of understory trees and adding hydraulic traits. This type of model that incorporates diverse tree strategies, detailed forest structure and realistic physiology is necessary if we are to be able to simulate tropical forest responses to global change scenarios.

2 useful files:

1. all_categories.txt: 34 lines (33 categories + "Unknown"), each line contains a category name.
2. asin2category.json: A mapping between parent_asin (item ID) to its corresponding category name.

This is a large-scale Amazon Reviews dataset, collected in 2023 by McAuley Lab, and it includes rich features such as:

- User Reviews (ratings, text, helpfulness votes, etc.); - Item Metadata (descriptions, price, raw image, etc.); - Links (user-item / bought together graphs).

What's New? In the Amazon Reviews'23, we provide:

Larger Dataset: We collected 571.54M reviews, **245.2% **larger than the last version; - Newer Interactions: Current interactions range from May. 1996 to Sep. 2023; Richer Metadata: More descriptive features in item metadata; Fine-grained Timestamp: Interaction timestamp at the second or finer level; Cleaner Processing: Cleaner item metadata than previous versions; Standard Splitting: Standard data splits to encourage RecSys benchmarking.

This data set provides annual spatial patterns of cropland, natural pasture, and planted pasture land uses across Amazonia for the period 1940/1950-1995. Two series of 5-minute grid cell historical maps were generated starting from land use classification products for 1995. Annual data are the fraction of natural pasture, planted pasture, and cropland in each 5-min grid cell. The annual maps are provided in two NetCDF (.nc) format file at 5-minute resolution. The AMZ-C.nc file covers the Brazilian portion of Amazon and Tocantins Rivers basins, and is based on the 1995 land use classification of Cardille et al. (2002), generated through the fusion of remote sensing (AVHRR) and agricultural census data. The second file, AMZ-R.nc, covers the entire Legal Amazon region and adjacent areas and is based on the 1995 land use classification by Ramankutty et al. (2008). The land use classification was generated by the fusion of satellite imagery (MODIS and VEGETATION-SPOT) and data from the agricultural census. A historical land-use reconstruction algorithm was used to generate the annual spatial patterns (based on work from Ramunkutty and Foley, 1999).

The dataset contains cloud network performance data related to the Amazon S3 storage service. The dataset refers to experimental campaigns conducted in May 2016. The dataset was collected leveraging 77 Bismark VPs, instructed as detailed in the following. Each VP performed repeated download cycles over 7 days. Each cycle is composed of 40 sequential download requests spaced out by 10 seconds and uniquely identified by a combination of factors, i.e. cloud region, file size, and storage class. Downloads within cycles are randomly scheduled and repeated from each VP every 2 hours. After every download, VPs run TCP-traceroute towards the IP address that served the request in order to trace the information related to the path and estimate the RTT to the S3 cloud datacenter (note that this information is not always available, due to the version of the firmware of the Bismark nodes and to the measurement tools available on them).

When refering to our data set, please cite the following reference: Valerio Persico, Antonio Montieri, Antonio Pescapè: On the Network Performance of Amazon S3 Cloud-Storage Service. CloudNet 2016: 113-118

In 2024, Aldi was the second-most favorite retailer among Amazon Prime users living in the United States. The net promoter score (NPS) they assigned to the grocery retailer was ****. Unsurprisingly, U.S. Prime shoppers ranked Amazon as their favorite retailer, with a net promoter score of **.

SARS-CoV-2 spread rapidly in the Brazilian Amazon. Mortality was elevated, despite the young population, with the health services and cemeteries overwhelmed. The attack rate in this region is an estimate of the final epidemic size in an unmitigated epidemic. Here we show that by June, one month after the epidemic peak in Manaus, capital of the Amazonas state, 44% of the population had detectable IgG antibodies. This equates to a cumulative incidence of 52% after correcting for the false-negative rate of the test. Further correcting for the effect of antibody waning we estimate that the final attack rate was 66%. This is higher than seen in other settings, but lower than the predicted final size for an unmitigated epidemic in a homogeneously mixed population. This discrepancy may be accounted for by population structure as well as some limited physical distancing and non-pharmaceutical measures adopted in the city.