As I wanted to retrieve all relevant genetic annotation for human genome using AnnotationDbi package as well as reference database org.Hs.eg.db, I have noticed following seemingly odd observation.

NCBI_gene_annotationDbi <- AnnotationDbi::select(org.Hs.eg.db, keys=AnnotationDbi::keys(org.Hs.eg.db, keytype="SYMBOL"), columns=c("SYMBOL", "GENENAME","ENTREZID"), keytype="SYMBOL")

duplicated_genes <- NCBI_gene_annotationDbi$SYMBOL[duplicated(NCBI_gene_annotationDbi$SYMBOL)]

NCBI_gene_annotationDbi_duplicated <- NCBI_gene_annotationDbi %>% dplyr::filter(SYMBOL %in% duplicated_genes)

As you can notice, only a small fraction (8 symbol over a total 61538) of SYMBOL have duplicated inputs in ENTREZID field. Besides, most of these duplicated values often integrate a general description and annotation of the gene, while the other ones seem superfluous (for instance, HBD has two ENTREZID matches, one general about hemoglobin subunit deltat, and another one really specific, and not describing its function: bone disease).

For such a small number of duplicated inputs, can't it be possible to clear all these seemingly superfluous one-to-many matches, only keeping for each the most relevant one. Or maybe merge both descriptions?

sessionInfo( )

R version 4.0.2 (2020-06-22)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: CentOS Linux 7 (Core)

Matrix products: default
BLAS:   /softhpc/R/4.0.2/lib64/R/lib/libRblas.so
LAPACK: /softhpc/R/4.0.2/lib64/R/lib/libRlapack.so

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C               LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8    LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C             LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats4    parallel  stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] GO.db_3.12.1              hgu133plus2probe_2.18.0   hgu133plus2.db_3.2.3      org.Hs.eg.db_3.12.0      
 [5] hgu133plus2cdf_2.18.0     bmkanalysis_1.0.0         testthat_3.0.1            affy_1.68.0              
 [9] EnsDb.Hsapiens.v86_2.99.0 ensembldb_2.14.0          AnnotationFilter_1.14.0   GenomicFeatures_1.42.1   
[13] AnnotationDbi_1.52.0      Biobase_2.50.0            GenomicRanges_1.42.0      GenomeInfoDb_1.26.2      
[17] IRanges_2.24.1            S4Vectors_0.28.1          BiocGenerics_0.36.0

The OrgDb packages are simply a re-packaging of existing data we get from NCBI (and some other places). We make no attempt to do anything more than ensuring that we get the current version of files from the annotation source and ensure that those data are parsed correctly and put into the underlying SQLite database. We don't have the ability to go through all the data we download and decide, on a case by case basis, whether or not there are superfluous matches.

Anyway, I wouldn't characterize much of what you present as superfluous matches. Although Biologists tend to rely on HUGO gene symbols, and think they are in some sense unique, they aren't. As evidenced by your list. I would be hard pressed to think that 'hemoglobin subunit delta' and 'hyphophosphatemic bone disease' (both HBD) are in any way the same thing.

Additionally, you should beware using more than one column for a search. The underlying query is a SQL inner join, and can balloon quite rapidly when you have any one-to-many matches.

The phenomenon you have discovered is well known. It's not a mistake, nor do most people find it a problem. You may find it inconvenient, but mammalian transcription is complicated. All gene annotation systems have incompatibilities and make unavoidable simplifications in an attempt to interpret an infinitely complex system within the framework of "genes". In this case you are comparing HUGO to Entrez. I suggest you try Ensembl, the other major system, then you might become more grateful for the simplicity and consistency of Entrez.

As I have remarked on this forum before, it is unfortunate that org.Hs.eg.db does not store the "Gene type" annotation provided by NCBI.

When I do expression analyses of human genes, I remove things that are not genes in the normal sense (Gene type=other or tRNA) and I restrict to reference chromosomes (removing Mitochondria and unassembled scaffolds). You will find that the redundancies between symbols and Entrez Ids then disappear.