Recommender systems in online stores rarely get clean data. Many products have no photo or no description, and the dislikes that users leave often go unused. I led two papers that each address one of these gaps with a small change to existing recommenders. Both were accepted at CIKM 2026, the ACM International Conference on Information and Knowledge Management, held in Rome from 7 to 11 November 2026.
SMD, short for Sequential Modality Dropout and written with Wenlong Zhang, is a four-line training change. When 95% of images and texts were removed at random, a recommender trained with SMD kept 61% of its accuracy, against 22% without it.
DualSpectralCF, written with Tong Qi of the University of Maryland and Xiaoxue Han of Stevens Institute of Technology, feeds dislikes into recommenders that need no training. On average, it ran 11 times faster than a trained model that also uses dislikes and reached 82% of its accuracy.
Both papers fall outside my dissertation. They test the same question it does, which is whether a small, cheap change can come close to a costlier one.

In this post
Why Recommenders Struggle With Real Product Data
SMD: Training for Missing Photos and Text
2.1 Hide a Whole Kind of Content During Training
2.2 Accuracy Held Up as Content Went Missing
2.3 Where SMD Stops
DualSpectralCF: Using Dislikes Without Training
3.1 A Dislike Counts as Half a Like
3.2 Better Than Its Base, and Much Faster Than a Trained Model
3.3 New Users Gain the Most
3.4 Where DualSpectralCF Stops
What It Means for Your Team
Questions Readers Ask
Read and Cite the Papers
1. Why Recommenders Struggle With Real Product Data
A recommender predicts what a user will want next from what that user did before. Many recommenders also read each product’s photo and text. Real catalogs have gaps in both, according to counts in the SMD paper:
In the public Amazon Reviews 2023 data, 34.3% of Toys & Games items and 48.3% of Beauty & Personal Care items have no text description.
In three of the four Amazon store categories used to test SMD, 25.7% to 40.9% of items have no image.
Most recommenders are trained and tested on data where every item has every kind of content. In the SMD experiments, a standard recommender kept only 40% to 73% of its accuracy when product text was missing at serving time, depending on the store category.
Dislikes are the second gap. Many platforms record negative signals, such as low ratings. The cheapest recommenders, which need no training, use only positive signals such as purchases and likes. The recommenders that do use dislikes must be trained, which costs far more computing time.
2. SMD: Training for Missing Photos and Text
2.1 Hide a Whole Kind of Content During Training
SMD changes how a recommender is trained and leaves the model itself as it is. It works with any recommender that combines an item’s ID, photo, and text in one step. The procedure has four steps:
For each user in a training batch, flip two coins, one for photos and one for text, each with a 30% chance.
If the photo coin comes up, hide every photo in that user’s history. The text coin works the same way.
Train with the same settings as before.
Serve the trained model unchanged, with nothing hidden and no added cost.
SMD hides content for a user’s whole history at once, because real gaps tend to cover a whole category or data feed. The change takes four lines of code. The authors note that random masking of content is an established training tool in other fields and do not claim it as new. What they add is masking whole user histories in sequential recommendation, the task of predicting a user’s next item from the order of past items. They also test it on four different recommenders.

2.2 Accuracy Held Up as Content Went Missing
The paper measures accuracy with HR@10, the share of users whose true next item appears in the recommender’s top 10 picks out of the whole catalog, where higher is better. Retention is the share of full accuracy that a recommender keeps when content is missing, also higher is better. The main recommender in the tests is MM-SASRec, a standard self-attention model that adds photo and text features to each item.
With all text removed in the Amazon Scientific category, MM-SASRec kept 83% of its accuracy with SMD and 40% without. The share of users whose next item reached the top 10 rose from 2.45% to 5.22%.
On four different recommenders in that category, retention with text removed rose in every case. The largest gain was for IISAN, from 18% to 56%.
When each photo and each text was removed at random with a 95% chance, MM-SASRec kept 61% of its accuracy with SMD and 22% without.
With all content present, accuracy changed by +0.8% on average across 11 combinations of recommender and store category. Six of the 11 lost some accuracy, by at most 4.4%.
In 22 of 24 per-user statistical tests, the difference between training with and without SMD was statistically significant, meaning unlikely to come from chance.

2.3 Where SMD Stops
SMD has these limits:
It covers only photos and text that are missing, not content that is wrong or of other kinds such as video.
The 30% chance was compared only with 50%, on one recommender.
Of the 16 possible combinations of four recommenders and four store categories, 11 were run, because of limited computing time.
The largest gains in robustness came with the largest losses in full-content accuracy.
All results come from offline benchmarks. No live test in a store is reported.
3. DualSpectralCF: Using Dislikes Without Training
3.1 A Dislike Counts as Half a Like
DualSpectralCF builds on spectral collaborative filtering, a family of recommenders that needs no training. These methods treat a user’s history as a signal spread over a graph of item similarity. A fixed formula smooths that signal and keeps the patterns that many users share. ChebyCF, GF-CF, and Turbo-CF are three such recommenders. They are fast, but they ignore dislikes. DualSpectralCF adds two settings to any of them:
Each disliked item enters the user’s history with its own weight, next to liked items at +1.
Items that users often dislike together get an extra link in the item graph, and the second setting controls its strength.
The main surprise concerns the first setting. The best weight for a dislike was never negative, on any recommender or dataset. At the default, a disliked item counts as half a like. The authors explain that a one-star rating of a horror film still shows an interest in horror. Treating dislikes as true negatives did worse than ignoring them in every case tested.

3.2 Better Than Its Base, and Much Faster Than a Trained Model
The tests use five public datasets with both likes and dislikes, released with SIGformer, a trained recommender that uses dislikes. Two come from Amazon, one from the review site Epinions, and two from the short-video app Kuaishou. The paper measures accuracy with Recall@20: of the items a user liked in held-out test data, the share that appear in the recommender’s top 20 picks, where higher is better. All changes below are relative to the base recommender’s score.
Each of the three DualSpectralCF versions matched or beat its base recommender on all five datasets.
At the default setting, the version built on ChebyCF gained 3.3% to 16.0% on four datasets. On the fifth, KuaiRec, the default hurt. KuaiRec labels a video as disliked from short viewing time, and there the best setting gave dislikes no weight and gained 1.9%.
The largest gain was 32.6%, for the version built on Turbo-CF on the KuaiRand dataset.
The ChebyCF version beat LightGCN, a trained graph recommender, on four of five datasets.
Against SIGformer, the ChebyCF version reached 71% to 91% of its Recall@20. It ran 7.7 to 155 times faster per dataset, 11 times on average, on one GPU.
The fastest version, built on GF-CF, averaged 8.5 seconds per dataset against SIGformer’s 270 seconds. Using dislikes added 16% to 59% to the run time of the base recommender.

3.3 New Users Gain the Most
Dislikes helped most for users with little history. For users with 1 to 5 items in their training history, the ChebyCF version raised Recall@20 by 2.2% to 29.2%, depending on the dataset. For the most active users, with 100 or more items, the change turned negative, a drop of up to 11.1% on KuaiRec.
3.4 Where DualSpectralCF Stops
DualSpectralCF has these limits:
The two settings are global constants. One weight for all users can penalize active users whose likes and dislikes span different topics.
On noisy dislike data, such as KuaiRec’s, the default setting hurt.
It still trails SIGformer, at 71% to 91% of its accuracy.
All results come from offline benchmarks. No live test is reported.
4. What It Means for Your Team
Both papers point to practices for teams that run recommenders:
Test your recommender with photos or text removed, and measure how much accuracy it keeps.
If your catalog has gaps, try training with whole histories masked. It costs one retraining and nothing at serving time.
Keep dislikes in your logs, and try them as weak positive signals before treating them as negatives.
Before paying for a trained model, measure a training-free recommender. DualSpectralCF reached 71% to 91% of a trained model’s accuracy in a small share of its time.
Check results by user group. Dislikes helped new users most and hurt the most active ones on some datasets.
5. Questions Readers Ask
How can a recommender keep working when product photos or descriptions are missing?
One option is to train it with that content hidden. SMD hides all photos or all text for a user’s whole history at random during training. With 95% of images and texts missing at random, a recommender trained this way kept 61% of its accuracy, against 22% without.
Can a recommender use thumbs-down ratings without training a model?
Yes. DualSpectralCF adds dislikes to recommenders that need no training. It matched or beat its base recommender on all five benchmarks tested and ran 11 times faster on average than SIGformer, a trained model that uses dislikes.
Should a recommender treat a disliked item as a negative signal?
In these experiments, it should not. The best weight for a dislike was never negative, and treating dislikes as true negatives did worse than ignoring them. A dislike still shows interest in a topic.
Is the code available?
Yes. The code for SMD and for DualSpectralCF is on GitHub.
6. Read and Cite the Papers
Guanqun Yang and Wenlong Zhang. “Sequential Modality Dropout for Robust Multi-Modal Sequential Recommendation.” Accepted at CIKM 2026. arXiv, code.
Guanqun Yang, Tong Qi, and Xiaoxue Han. “DualSpectralCF: Training-Free Sign-Aware Spectral Collaborative Filtering.” Accepted at CIKM 2026. arXiv, code.
The overview post places both papers next to my dissertation work on AI for software testing and security.
To cite the papers:
@misc{yang2026sequentialmodalitydropoutrobust,
title = {Sequential Modality Dropout for Robust Multi-Modal Sequential Recommendation},
author = {Guanqun Yang and Wenlong Zhang},
year = {2026},
eprint = {2608.10240},
archivePrefix = {arXiv},
primaryClass = {cs.IR},
note = {Accepted at CIKM 2026},
url = {https://arxiv.org/abs/2608.10240}
}
@misc{yang2026dualspectralcftrainingfreesignawarespectral,
title = {DualSpectralCF: Training-Free Sign-Aware Spectral Collaborative Filtering},
author = {Guanqun Yang and Tong Qi and Xiaoxue Han},
year = {2026},
eprint = {2608.10247},
archivePrefix = {arXiv},
primaryClass = {cs.IR},
note = {Accepted at CIKM 2026},
url = {https://arxiv.org/abs/2608.10247}
}
Guanqun Yang, LLM Researcher. More work at guanqun-yang.com and on Google Scholar.

